H2020 Marie Skłodowska-Curie Actions Artificial Intelligence ...

H2020 Marie Skłodowska-Curie Actions

Artificial Intelligence Cluster Meeting

Research Executive Agency

Meeting Report and Key Messages for Policy Consideration

Horizon 2020 MSCA Artificial Intelligence Cluster Meeting

2

RESEARCH EXECUTIVE AGENCY

Excellent Science Department

Contact: REA INFO

E-mail: [email protected]

European Commission B-1049 Brussels

© European Union, 2020

The reuse policy of European Commission documents is regulated by Decision 2011/833/EU (OJ L 330, 14.12.2011, p.39). For any use or reproduction of photos or other material that is not under the EU copyright, permission must be sought directly

from the copyright holders.


3

H2020 Marie Skłodowska-Curie

Actions

Artificial Intelligence

Cluster Meeting

Meeting Report

and

Key Messages for Policy Consideration


4


5

Table of Contents

1. Background ........................................................................... 11

1.1 What is Artificial Intelligence ......................................................................................... 11

1.2 MSCA and Artificial Intelligence .................................................................................... 12

1.3 Meeting concept ...................................................................................................................... 15

2. Policy considerations ................................................................ 17

2.1 Identified needs ...................................................................................................................... 17

2.2 Future directions .................................................................................................................... 19

3. Artificial Intelligence research in MSCA .................................... 20

3.1 AI in MSCA DIGITAL projects .......................................................................................... 21

3.2 AI in MSCA HEALTH projects ........................................................................................... 27

3.3 AI in MSCA ENVIRONMENT projects ........................................................................... 31

3.4 Cross-cutting challenges for AI research ................................................................ 34

4. Meeting conclusions ................................................................. 39

References ................................................................................... 42

Annexes ....................................................................................... 43


6

ACKNOWLEDGEMENTS

This report is based on the work of the three experts invited to take part in the event and appointed to draft the meeting report: Dr Ariana POLYVIOU,

Lecturer in Information Systems Management at University of Nicosia (Cyprus), and in charge of leading this report; Dr Marusela OLIVERAS

SALVA, Intellectual Property and Innovation expert; and Dr Alain VERSCHOREN, Professor in Mathematics and Computer Science at the

University of Antwerp (Belgium), who also delivered the keynote presentation which set the scene for the event.

The experts’ contribution is mainly presented in chapters 2 and 3 of the report, whereas the introduction and conclusions (chapters 1 and 4) have

been drafted by the editorial team of the Research Executive Agency Marie Skłodowska-Curie Actions operational units who were also in charge of the

organisation of the event: Ioanna PEPPA (REA.A1), Alina SUHETZKI

(REA.A2), Amanda Jane OZIN-HOFSAESS (REA.A3), Irina TIRON (REA.A3) and Maria CARVALHO DIAS (REA.A4), under the supervision of Fredrik OLSSON HECTOR (REA.A3).

Disclaimer

Neither the European Commission nor any person acting on behalf of the

Commission is responsible for the use that might be made of this document. The contents presented in chapter two and three of this

document are the sole responsibility of the experts appointed for this event. Although REA staff facilitated the preparation of this document, the

views expressed herein reflect the opinion of the experts and may not in any circumstances be regarded as reflecting an official position of the European Commission.


7

ACRONYMS

AI Artificial Intelligence

COFUND Co-funding of Regional, National and International Programmes

DG CNECT Directorate General for Communications Networks, Content and Technology

DG EAC Directorate General for Education, Youth, Sport and Culture

DG ENV Directorate General for Environment

DG GROW Directorate General for Internal Market, Industry, Entrepreneurship and SMEs

DG HOME Directorate General for Migration and Home Affairs

DG JUST Directorate General for Justice and Consumers

DG R&I Directorate General for Research and Innovation

EU European Union

HLEG European Commission’s High-Level Expert Group in AI

IF Individual Fellowships

ITN Innovative Training Networks

JRC Joint Research Centre

MSCA Marie Skłodowska-Curie Actions

NIGHT European Researcher’s Night

REA Research Executive Agency

RISE Research and Innovation Staff Exchange

SME Small and Medium-sized Enterprises


8

FOREWORD

Artificial Intelligence (AI) has been and remains at the top of the European

Commission’s agenda. Many initiatives have been launched to shape a European approach to AI and thus ensure that the EU is on top of technological developments, encourages uptake by the public and private sectors, and prepares

itself for the socio-economic changes that will be brought about by AI, while also establishing an appropriate ethical and legal framework.

Part of Horizon 2020’s Excellent Science pillar, the Marie Skłodowska-Curie

Actions (MSCA) support researchers in all scientific domains and at any stage of their careers. The world-class talented researchers and their AI research took the

central stage at an event jointly organised by the MSCA units of the Research Executive Agency (REA) and the Directorate General for Education, Youth, Sport, and Culture (DG EAC) on 10 and 11 December 2019 in Brussels. Over one and a

half days, the invited researchers presented the contribution of their respective research projects to the development of the state-of-the-art in AI. They also

engaged in fruitful exchanges with Policy and Project Officers about the AI applications in the digital, health and environment sectors. The event demonstrated the high potential of the Marie Skłodowska-Curie researchers to

actively contribute to the development of current and future policies in different scientific fields. We are thankful to the very enthusiastic and committed

researchers and coordinators who accepted to join this event and share their work. We are also thankful to the Policy Officers who helped align this event with the European Commission policy agenda and who kindly accepted to give

presentations and lead the discussion as moderators on behalf of the Directorates General for Research and Innovation (DG R&I), Migration and Home Affairs (DG

HOME), Communications Networks, Content and Technology (DG CNECT), Justice and Consumers (DG JUST), Environment (DG ENV), Internal Market, Industry, Entrepreneurship and SMEs (DG GROW) and the Joint Research Centre (JRC). We

are also grateful to the scientific and innovation experts who attended the event and helped with the meeting report.

This report is mainly addressed to those European Commission’s services

interested in learning about the contribution of MSCA researchers to the latest AI research, but also to the Research and Innovation community more broadly. By

joining forces in organising this event we are convinced that the REA and the European Commission’s policy representatives created further synergies, stronger capacity, and more shared value, thus improving policy uptake, as well as overall

collaboration. We hope that this event will pave the way for other similar events tackling related topics of interest to those in Europe and beyond.

Alessandra LUCHETTI Head of Department, REA, Excellent Science

Claire MOREL Head of Unit, DG EAC, Unit C2 MSCA

https://ec.europa.eu/digital-single-market/en/artificial-intelligence#A-European-approach-to-Artificial-Intelligence

https://ec.europa.eu/digital-single-market/en/artificial-intelligence#A-European-approach-to-Artificial-Intelligence


9

EXECUTIVE SUMMARY

Artificial Intelligence (AI) impacts Research and Innovation (R&I) in many fields

of knowledge and has direct consequences on our daily lives. Technology is moving forward at a rapid pace and getting a grasp of the scope and application of AI research is not easy. Moreover, paving the way forward to ensure AI

applications are reliable and trustworthy, while also safeguarding people from the misuse of technology, is also challenging.

AI has been and remains a priority on the European Commission’s agenda. In this

light and in support of the joint policy making approach, the Research Executive Agency (REA), together with the Directorate General of Education, Youth, Culture,

and Sport (DG EAC) organised a thematic meeting of projects from the Marie Skłodowska-Curie Actions (MSCA) portfolio contributing to the EU-funded research on AI. The objective of this cluster event was three-fold. First, to

showcase MSCA projects and their contribution to European AI research in the digital, health and environment fields and hence to understand the R&I state-of-

the-art and the future trends. Second, to foster new contacts and collaborations between scientists and to help generate new ideas in their scientific disciplines. And finally, to promote face-to-face discussion between research practitioners

and policy makers to raise awareness of the challenges of AI research and to outline future policy needs.

Under Horizon 2020, the MSCA programme alone has supported around 10.000

projects across all R&I domains. More than 300 grants signed in the MSCA calls between 2014 and 2018 focus on AI and its applications. Trends in applications for this bottom-up programme suggest that projects dealing with AI applications

will increase in the future, both in quantity and in diversity, especially given their strong multidisciplinary nature.

With such a diverse portfolio of projects, a working AI definition was suggested

for this meeting referring to the design and action of a system in response to a perceived signal, and a degree of autonomy of the system (see section 1.1 for

further explanation). The definition is a good basis for understanding the key points presented by the researchers and for following the discussions led by the Policy Officers. Nevertheless, the researchers involved in the meeting agreed

that, given the rapidly evolving AI concept, any AI definition easily reaches its limitations since it hardly covers the questions raised. For this meeting, the

experts focused on describing and analysing the specific topics and fields where AI is dynamically present, in order to help both policy makers and researchers better explain the applications used (e.g. Machine Learning, Machine Reasoning,

Robotics), as well as identify the main challenges.

The invited researchers, policy makers and experts also discussed the future directions AI research is expected to take and the policies needed to ensure that:


10

(1) researchers can perform their work (e.g. data sharing, data access, data

protection) in an effective and responsible manner; (2) the research outcomes lead to trustworthy and reliable AI applications;

(3) the applications are understood, correctly implemented and accepted by the end-users from the wider society. For these points, a “foresight approach” was used to tackle questions in the panel sessions.

The main findings of the meeting have been grouped into six main points that the scientific and innovation experts presented in the core chapters of this report.

They are the result of a needs assessment and are addressed for policy development, for awareness raising, and for the present and future training of

researchers:

1. Need for interdisciplinarity 2. Need for training

3. Need for governance (ethical and legal issues) 4. Need for data access

5. Need for innovation management 6. Need for good and reliable communication

Nurturing the flow of information and fostering synergies between the various initiatives at the EU and global level is key for effective joint policymaking. Such thematic cluster meetings on topics relevant for the EU and beyond (e.g. AI,

cancer, climate change, healthy food, etc.) are a good means of communication and collaboration with a main goal of facilitating interaction between policy

makers and research practitioners.


11

1. Background

In April 2018, the European Commission presented for the first time an AI

strategy and agreed on a coordinated plan with Member States. The High-Level Expert Group (HLEG) on AI presented its Ethics Guidelines on trustworthy AI in

April 2019. Moreover, the new European Commission has been entrusted with coordinating the work to further define the European approach on trustworthy AI, including its human and ethical implications. In her Political Guidelines,

Commission President VON DER LEYEN presented the plan for a Europe fit for the digital age, where ethical AI and the use of big data to create wealth for societies

and businesses are core topics.1

The present report provides an overview of the keynote presentation, project presentations, panel discussions, poster and networking sessions of the MSCA AI

meeting that was organised on 10 and 11 December 2019 in Brussels.2 The main objectives of the report are: to provide an overview of the current state-of-the-art in research on AI that is being conducted by the MSCA projects that

participated in the event; and to summarise and consolidate the discussions from each panel into a set of six main considerations for policy makers.

1.1 What is Artificial Intelligence3

Professor Alain VERSCHOREN, the keynote speaker who opened the event,

suggested the following working definition of AI, with emphasis on a core element, i.e. the degree of autonomy of the system, to not only propose solutions

but to also take decisions:4

AI refers to systems that, given a complex goal, act in the physical or digital world by perceiving their environment, interpreting the collected structured or

unstructured data, reasoning on the knowledge derived from these data and proposing the best action(s) to take (according to predefined parameters) to

achieve the given goal.

1 See latest White Paper on Artificial Intelligence: a European approach to excellence and trust and

Factsheet: Excellence and Trust in Artificial Intelligence (published online in February 2020). 2 The agenda, background material, photos, panel and poster presentations are available on the event’s webpage. We thank the participants who kindly agreed to share their presentations. 3 This section re-uses fragments from the keynote presentation entitled De-mystifying Artificial Intelligence, delivered by Professor Alain VERSCHOREN from University of Antwerp. 4 For other definitions of AI see the following EU references: JRC report Artificial Intelligence: a

European perspective, or the report by the European Commission’s HLEG on AI, A definition of Artificial Intelligence: main capabilities and scientific disciplines.

https://ec.europa.eu/digital-single-market/en/news/commission-presents-plan-make-most-artificial-intelligence

https://ec.europa.eu/digital-single-market/en/news/commission-presents-plan-make-most-artificial-intelligence

https://ec.europa.eu/commission/presscorner/detail/en/IP_18_6689

https://ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai

https://ec.europa.eu/commission/commissioners/2019-2024/vestager_en

https://ec.europa.eu/info/strategy/priorities-2019-2024_en

https://ec.europa.eu/info/strategy/priorities-2019-2024/europe-fit-digital-age/excellence-trust-artificial-intelligence_en

https://ec.europa.eu/info/horizon-2020-msca-cluster-event-artificial-intelligence-10-11-december-2019-brussels_en

https://ec.europa.eu/jrc/en/publication/artificial-intelligence-european-perspective


https://ec.europa.eu/digital-single-market/en/news/definition-artificial-intelligence-main-capabilities-and-scientific-disciplines



12

Although this definition is reasonable, it has its own limitations. It makes more sense to just describe what kind of topics AI considers, studies and applies. In fact, as a scientific discipline, AI deals with numerous approaches and techniques

such as Machine Learning (of which deep learning and reinforcement learning are specific examples), Machine Reasoning (which includes planning, scheduling,

knowledge representation and reasoning, search, and optimisation), Robotics (which includes control, perception, or sensors), etc. It is important to perceive AI as a tool.

Just as not all that happens on the internet favours humanity, some fail-safes have to be built into the realm of AI. It is essential that blackbox algorithms produced by artificial systems are scrutinised carefully. There is no problem with

a system developing strong algorithms for the game of chess, but things are quite different if uncontrolled algorithms have to make ethical choices or

decisions, give political advice, or help companies hire people on the basis of certain parameters.

Europe has developed clear strategies and programmes to guide AI development,

with a shared concern for an agreed ethics and security framework and applications, with the goal of benefitting European society and strengthening European values.5 European AI is thus strongly human-centred and could/should

serve as an example for the international community. More than 30% of worldwide papers on AI come from Europe, ahead of China, and with only the US

slightly stronger (with 33%).6

There is a strong need for regulations and a clear legal framework concerning ownership and sharing of data among stakeholders (e.g. public sector, industry,

users, and developers). Policy makers should also proactively consider the impact of AI on teaching, learning, education and, in particular, the overall impact on human intelligence and cognitive capacities, on the relationship between

education, work and human development, the impact on medicine and health (e.g. robotics, diagnostics, personalised medicine), as well as on the environment

and many other fields.

1.2 MSCA and Artificial Intelligence

Under Horizon 2020 (2014-2020), the MSCA programme has provided 6.2 billion euro to support 65.000 researchers across Europe and beyond.7 The programme

5 See the reports of European Commission’s HLEG on AI: Ethics guidelines for trustworthy AI

(2019), or Policy and investment recommendations for trustworthy Artificial Intelligence (2019). 6Artificial Intelligence: a European perspective, report by JRC (2018). 7 Since 1994, the programme has trained around 130.000 researchers – including 35.000 doctoral candidates - throughout their career. By the end of 2020, the MSCA will have supported a total of


https://ec.europa.eu/digital-single-market/en/news/policy-and-investment-recommendations-trustworthy-artificial-intelligence



13

is about achieving excellence in science - as part of the Horizon 2020 strategic

pillar Excellent Science - and excellence in research careers. Both elements are important in fostering a European Research Area and are important contributions

to realising the vision of a European Knowledge Area from east to west. The MSCA provide grants for all stages of researchers’ careers and support the development of knowledge and enhancement of skills through mobility; mobility

in the sense of crossing borders, by working in countries in Europe and around the world; crossing sectors by performing research and training in both academic

and non-academic organisations; and crossing disciplines by doing research that requires collaboration between different scientific disciplines. The MSCA enable research-focused organisations to host talented researchers and provide them

with innovative research training, and by also creating worldwide strategic partnerships with leading institutions. The actions are open to all domains of R&I,

from fundamental research to market take-up and innovation services, with a strong participation of industry (large industrial concerns as well as SMEs).

Scientific fields are chosen freely by the applicants in a fully bottom-up manner. Moreover, though the high-quality work of the funded researchers and their multidisciplinary approach, the MSCA shed light on the latest trends in innovative

research. Below is a summary of all MSCA and their main characteristics (Figure 1):

Figure 1: Overview of MSCA

145.000 researchers, including almost 40.000 PhD candidates. For more statistics see Marie

Skłodowska Curie actions - Facts & Figures or the factsheet Marie Skłodowska-Curie Actions - Driving

innovation, cultivating excellence in doctoral and postdoctoral training, both published in 2019.

https://ec.europa.eu/research/mariecurieactions/resources/document-libraries/marie-sk%C5%82odowska-curie-actions-facts-figures_en


https://ec.europa.eu/research/mariecurieactions/resources/document-libraries/factsheet-msca-audience_en



14

In addition to the four actions listed above, the European Researchers’ Night

(NIGHT) supports Europe-wide public events aimed at bringing researchers closer to the public and stimulating interest in research careers, especially among young

people.

The MSCA have so far funded more than 9.500 projects across all R&I domains. For the MSCA AI meeting more than 7.000 projects – representing advanced and

completed projects from the MSCA calls 2014-2018 – were considered (further details on analysis below). The analysis showed that 312 projects were dealing

with AI and AI-related applications (see breakdown in Figure 2) in various areas such as agriculture, architecture, cultural heritage, cyber-security, digital communication, digital identity, disaster management, e-government, energy

efficiency, environment, resources and sustainability, health, industrial technologies, intelligent robotics, cybernetics, the Internet of Things, the

management of natural disasters, transport, etc. As a bottom-up programme, one of the strengths of the MSCA is that it enables researchers to contribute to joint policy-making through their high-quality results and fresh ideas. Moreover,

given the multidisciplinary nature of the research, AI-based projects are expected to significantly increase in the future, both in quantity and in the domains

covered.

Figure 2: Breakdown of AI funded projects in 2014-2018 MSCA calls 8

8 The analysis was made for the pupose of the meeting and took into account advanced and closed projects from all MSCA calls from 2014 to 2018 (included). The breakdown per MSCA reflect the

mapping process undergone by the MSCA operational units. See below section Methodology for selection of MSCA AI projects for more details.


15

1.3 Meeting concept

Scope

REA manages a large portfolio of H2020 projects and one of its roles is to provide programme and policy feedback on the funded actions to the European Commission. REA’s four MSCA units joined forces with DG EAC to organise a joint

project cluster event on AI. The MSCA units partnered with MSCA researchers and and policy makers in the field to provide the participants with an interesting

scientific programme and a series of panel discussions on policy areas that could affect our lives in the digital world of the future, both in health and in the environment. The event aimed at:

showcasing the contribution of MSCA projects to AI R&I;

promoting discussion and collecting information to provide coordinated input to the relevant EU policy-making services;

enhancing synergies among projects and creating or reinforcing networking

opportunities, particularly for MSCA fellows.

Over 150 participants were invited, including project coordinators, researchers, SME representatives, Project Officers and Policy Officers. Eight European

Commission services contributed to the event, namely DG EAC, DG R&I, DG HOME, DG CNECT, DG ENV, DG JUST, DG GROW and the JRC.

Methodology for selection of MSCA AI projects

An important step that shaped the event was the mapping of the MSCA portfolio

to identify the relevant AI projects among the over 7.000 projects considered. In a first step, the projects were screened using AI-related keywords such as

artificial intelligence, machine learning, deep learning, reinforcement learning, machine reasoning, artificial neural networks, robots/robotics, human computer interaction, control systems, support decision, autonomous, computer vision etc.

This was followed by an in-depth analysis of the AI portfolio established for each MSCA, with a total of 312 AI projects identified. Given their recurrence and their

relevance for the European Commission policy agenda and to better focus the meeting, three policy domains were selected for this event, namely digital, health and environment.

Each of the four units involved in the organisation of the event undertook a qualitative assessment of the AI projects to select the most successful projects in the chosen domains. The Project Officers were key players in this process. As a

result, 45 projects were invited to the event, both ongoing and completed projects, of which 37 expressed their interest and availability to take part in six

panel discussions and poster sessions relevant for both scientific and policy aspects of AI.


16

Meeting format and content

A plenary session format was selected for the thematic panels, offering researchers specialised in computer programming the possibility to interact and to examine the opportunity of applying their algorithms in various environments.

Thus, the three selected policy domains were organised in five scientific panels: two in digital, two in health and one in environment. Each scientific panel

included speakers from four successful projects selected to represent each of the MSCA. These panels were moderated by Policy Officers from different EC Directorate Generals and were introduced by management of the REA MSCA

operational units.

The first panel explored AI applications in the digital world. The four projects provided insights on the control of swarm UAVs for localisation and tracking; how

emerging methodologies such as blockchain could be used with the purpose of securing transactions between UAVs; how an artificial vision platform could

contribute to the future sensing systems applied in security, health, marketing and navigation aids for driverless systems; and on the use of large scale measurements and machine learning to protect users in the social web.

The second digital panel showcased results for mobile-based reminding solutions;

activity recognition and behaviour modelling as support for people with dementia; and rapid skill learning in complex robots acting in a real environment. Moreover,

the invited speakers also discussed advances in the science and technology required to realize meta-cognitive functions in social robotics and new solutions for decision-making support in the buying and selling of electricity.

The agenda also included two panels that covered AI applications in the health sector. The invited researchers presented their advancements on: medical imaging and simulation models to support decision making on treatments and

other medical interventions; a novel software which will help to improve the clinical protocols for dosing of therapeutic drugs in children; a model using robotic

devices together with perceptual classifiers utilising deep learning to help autistic children learn behavioural cues about emotional affects and artificial adaptive learning systems embedding the cerebellar neural network; and machine learning

mechanisms with the aim to develop a modular biomimetic architecture for controlling robots and to get insights into the brain capability to achieve motor

adaptation and learning.

Other AI health projects showed results for a microwave technology that can identify malignant tumours in real-time, or solutions around the detection, assessment and treatment of pathological speech. For people living with diabetes

new technologies in DIY APS were revealed aimed at obtaining improved user experience, as well as improved predictive capacity. Moroever, another

researcher presented an automated tool for processing of large-scale social media data in order to understand patient feedback.


17

The panel on AI applications in the environment field started by emphasising the

importance of the Green Deal, a major strategy launched by the European Commission on the same day. The researchers presented novel multi-agent deep

reinforcement learning algorithms for enabling the coordination of multiple autonomous drones for forest firefighting; novel neurocognitive VLSI multi-sensor architecture for the electrochemical analysis of fluids in real time; a novel IoT

smart framework; integrated and intelligent data management for smart sustainable urban water environments; and a computerised tool and prototypes

for a novel Computer and Data Centre (CDC) dew point cooling system.

Alongside the 20 MSCA projects contributing to the scientific panels, a further 17 projects were invited to present their work through posters. A dedicated session

for project posters was included in the agenda and all the networking activities occurred in the room where the posters were exhibited in order to increase their visibility and encourage the sharing of experience among the participants.

Furthermore, the meeting also included presentations by EC representatives

working in AI-related programmes. The European strategy on AI and the AI funding opportunities in H2020 and Horizon Europe were presented, including

initiatives such as AI-on-Demand-Platform, Digital Innovation Hubs and European AI Alliance. Moreover, one panel was dedicated to cross-cutting topics such as ethics and how the EU could position itself as frontrunner in setting trustworthy

and ethical by-design AI. Presentations covered topics of AI and education, ethics guidelines proposed by the HLEG, and AI in EU security research.

The appointed experts actively contributed to the discussions with the

participants both in the panel and networking sessions and provided a foresight view in the concluding session. The experts provided the take-away messages for

policy considerations as well as further recommendations (chapter 2 of this report). They also provided a robust summary of the presentations and panel discussions (chapter 3).

2. Policy considerations

The experts summarised the meeting by highlighting six key messages, which are

identified as “needs” for policy consideration.

2.1 Identified needs

1 Need for interdisciplinarity: maximising investment in interdisciplinarity and the exchange of know-how is of core importance; the success of AI during the last decades has been strongly influenced by joint research across distinct disciplines. This has also helped to close the gap between, for example, computer

scientists working in AI and researchers in medicine, who frequently do not understand their mutual “language” and thus miss opportunities for cross-

discipline applications. MSCA projects as well as strong interactions with industry can play an important catalyst role here.


18

2 Need for training: researchers should be trained in using and understanding

the new possibilities that emerging AI could provide in their field, eliminating routine matters and thus leaving more time for creative “human” activities. One

might compare this “mind click” to the switch that pure mathematicians or medical doctors made several decades ago, when they accepted that computers could be very useful tools when properly integrated in their research.

3 Need for governance (ethics and regulation): whereas human beings have moral capacities, algorithms lack such capacities. AI will only become trustworthy

through the presence of sufficient ethical and other control mechanisms, especially when its intended results have a direct impact on humans.

4 Need for access to data: an AI system can only return high quality results,

decisions or advice when it is fed with a sufficiently broad and relevant set of data. A proper exchange of results between the different actors requires an open platform. However, the possible conflicts to be expected between open access to

training and other data versus privacy or other misuse should be regulated without further delay within a transparent legal framework.

5 Need for innovation management: the innovation aspects associated with

AI will be quite different from the traditional ones, in particular in view of the extreme change in paradigm and the risk of the presence of non-transparent

“black box” algorithms developed by intelligent systems. There will be a strong need for new proof of concept instruments and extended criteria to evaluate innovation potential, in particular when evaluating AI related MSCA proposals.

Emphasis should also be placed on the re-use and re-usability of previously obtained research results.

6 Need for good and reliable communication: the communication between scientists, policy-makers and the public is key for moving forward. Nevertheless, AI is not the solution to everything: one may only expect it to yield decent results

from well-structured data. AI could and should be applied where it is highly useful, for example as an educational or diagnostic tool, although in the near

future one may expect AI to be also used for decision-making in different areas (e.g. law, health care). Users should be informed about the intrinsic limitations of AI.

There is a natural fear of the unknown. AI is not well understood or indeed misunderstood. It is important to create ways of engaging and communicating with the public about the uses/misuses and future potential of AI. The public

must feel that AI is trustworthy, appropriately used, and ethically sound.

Following the experts’ analysis of the meeting’s outcomes, the findings were organised into strengths, weaknesses, opportunities and threats (SWOT), for

policy consideration, and are presented in the table below:


19

Table 1 MSCA AI projects: SWOT analysis

2.2 Future directions

AI represents a very wide family of techniques that can be useful for resolving

many practical problems in the future. From supporting stock market decisions to military services or social services, AI has the potential to provide significant

value to the wide spectrum of different fields in which it can be applied. Among these, the use of AI to develop socially useful applications is expected to be of vital importance. To this end, we recommend that priority be given to AI

applications that are expected to resolve important social problems in the fields of health, education and the environment.

Beyond using cutting-edge AI approaches in future projects, it is important to

ensure that projects deliver exploitable and useful results. Although employing AI approaches for developing parts of a problem are important, more focus should

be provided to projects that aim to tackle end-to-end system problems. In an attempt to identify targeted solutions to existing problems, AI-based solutions have the potential to also be applied to other areas and can thus be introduced

on the market and exploited as products. Moreover, issues relevant to the cost of the handler to deploy this kind of technology need to be considered. Future calls

could encourage projects to develop solutions that are cost-effective and potentially affordable by the target audience.

Experts consider that there is a need to strive for a better understanding of AI and for a greater openness to novelty and change. Much of the required scientific

knowledge to develop new and interesting AI applications is already available and

STRENGTHS WEAKNESSES OPPORTUNITIES THREATS

Improved

decision-making and automation

High cost of product Industrial partnerships

Cyber-security

Higher interaction

and integration

Lack of data

Regulatory framework

Privacy

Clustering data Lack of sufficient interdisciplinarity

Investment boosting capacity

Accountability

Personalisation and customer

orientation

Training

Ethics

Regulated access to

data


20

is rapidly expanding. Developers, engineers, and related computing scientists

know there is a need for a paradigm shift from focusing on technology to rethinking their research aims in terms of applications, skills of future end-users,

and the needed safeguards for society at large. Now is the time to engage in a dialogue between researchers, policy-makers and the public and to ensure that AI approaches are well selected, reliable and ethical.

AI and its use of data may fuel economic growth in Europe at several levels. AI technology has the potential to provide a vast amount of data from a wide range of interactions, helping improve decision-making in many applications and

sectors.

EU-funded research and the Horizon 2020 programme, as well as the upcoming Horizon Europe framework programme, can build on the identified strengths (see

Table 1). There are plenty of instruments available for funding, as well as initiatives to foster networks and turn ideas into innovations. As far as the weaknesses are concerned, these are areas where policy can play a role.

Particularly for the ethics and legal issues related to data access, the lack of data, and also financial barriers to production. As for opportunities, this is fertile ground

for policy actors and for scientists to get involved. Policy-makers and scientists need to work hand-in-hand to discuss, develop, pilot, and implement regulations in areas such as data access, data protection, ethics and communication. The

research community can encourage industrial partnerships and attract investment by taking advantage of existing initiatives and the available funding. In terms of

threats, it is clear that misuse of AI and concerns of ethics and accountability can slow down research in the EU.

Only by nurturing innovation in AI at national, regional and local levels will

Europe be able to continue to attract highly-skilled talent and support new initiatives. AI will create more jobs in this digital era. By providing high-quality training to researchers, the MSCA helps to avoid a brain drain, both outside

Europe and within it, such as from poorer regions to wealthier ones. The MSCA invests in (re-)skilling the European workforce by supporting the development of

their scientific and soft skills, expanding their professional network, and offering career development and training opportunities for the future.

3. Artificial Intelligence research in MSCA

Three scientific and innovation experts were appointed to attend the meeting and provide a full report of the event including a summary of the key elements of the

presentations in order to understand the state-of-the-art in applications of AI in the fields of the digital world, health and the environment. They captured the core points addressed in the panel discussions and provided their opinion on the

innovation potential in the field, as well as areas where policy-making is needed to meet the needs of scientific developments. All these areas are covered in the

current chapter.


21

3.1 AI in MSCA DIGITAL projects

Overview of project presentations

Although swarm robotics systems prompt a wide spectrum of opportunities for future applications, there is currently a lack of security standards in the field and

thus robotic systems may be deliberately hacked. The Blockchain: a new framework for swarm RObotic Systems (BROS) project combines swarm robotics systems and blockchain aiming to construct a new security behaviour for

distributed robotic systems. Blockchain is starting to demonstrate good potential to assist robotics and AI in becoming more secure. The project’s approach is

promising and unique, and aims to provide solutions to important security challenges in the field of robotics and AI operations. However, as noted during the panel discussions, although the approach enhances the ‘obstacles’ towards

security threats, it does not guarantee the complete elimination of the security risks. The BROS project is bridging the two different research areas aiming to use

robotics and blockchain to facilitate trustable autonomy. The fellow reflected on his experience at the MIT Media lab, highlighting the value of bringing researchers from different disciplines together.

The aim of the Advanced Hardware/Software Components for Integrated/Embedded Vision SystEms (ACHIEVE) project is to develop a distributed artificial vision platform composed of networked, smart and efficient

embedded vision systems. The project focuses on specific application scenarios of autonomous surveillance or intelligent transportation systems. The project

addresses the need for compact, lightweight and power-aware embedded vision systems, and suggests an approach for a distributed processing strategy.

Along these lines, and aiming to enhance safety and security in cities, the High-Accuracy Indoor Tracking and Augmented Sensing using Swarms of UAVs

(AirSens) project focuses on the use of UAVs (Unmanned Aerial Vehicles) for air-monitoring of cities. In particular, it focuses on the assessment of UAVs’ tracking

of performance, mapping capabilities, and distributed navigation control. Both ACHIEVE and AirSens involve monitoring of human activities and traffic management, and hence data handling, privacy and GDPR are some of their

major challenges. As identified during the panel discussion, in some cities, facial recognition has been banned, which may reduce the potential applicability,

adoption and impact of these technologies. Reflecting on this topic, one of the researchers highlighted that to address this issue their project follows a ‘privacy

by design’ approach such that the meta-data, rather than human identification data, are being handled by the system.

The EnhaNcing seCurity And privacy in the Social wEb: a user centered approach for the protection of minors (ENCASE) project focuses on the

protection of minors (especially adolescents) on online social networks. The consortium works on detecting and classifying malicious activity or sensitive

content (e.g., fake accounts, racist text, sexual predators, cyberbullies, etc.). Drawing on constantly updating machine-learning-based classifiers, the project’s

https://cordis.europa.eu/project/id/751615





22

mechanism remains effective as threats evolve. At the same time, ENCASE's

approach respects the privacy of adolescents and acts as a parental advice tool, rather than a control tool, with a focus on encouraging discreet supervision.

Similarly to other projects, the project results raise ethical issues relevant to the deep monitoring of the online social media activity of adolescents. As noted by the project speaker, it is necessary that the target subjects (i.e. adolescents)

authorise and accept parental control and the deep monitoring of their online activity.

Focusing on the integration of reminding technologies within the everyday lives of

people with dementia, the Use of computational techniques to Improve compliance to reminders within smart environments (REMIND) project is

seeking to develop an effective suite of reminding technologies in smart environments. In particular, the project uses AI and behavioural science to support the improvement of existing reminding solutions. The project involves the

home monitoring of people with dementia and thus sensitive private data are being produced and processed. To address this challenge, the project ensures

that the data is processed without leaving the house of the individual. One of the project’s innovations is a digital application for people with (pre-) dementia called MEMAS (MEMory ASsistant), a reminding, entertaining, memory training and

therapy app which is being exploited in the market. The app can be personalised and adapted to each individual user’s abilities and interests via the configuration

web-system. The project’s innovative outcomes also include a millimetre-wave radar for remote-sensing of vital signs for the contactless, live-streaming assessment of respiration patterns in people. This potential new product can be

introduced as new to the market and displays easily appreciated advantages to customers. In addition, the project has enabled the deployment of new consulting

services within one of the partner organisations in an emerging market where there is a growing demand and few offerings available with no major players. The REMIND project involves five industrial partners of which one is particularly

interested in adding the technological improvement under development to its product pipeline. The contractual arrangement among the beneficiaries allows this

partner to benefit from the outputs of the project, thereby facilitating the technology transfer process and fostering early adoption. The REMIND scientist-in-charge also acknowledged the involvement of its organisation in assisting with

the management of intellectual property assets and the technology transfer process.

The Applications of Personal Robotics for Interaction and Learning

(APRIL) project focuses on the design of companion robots that can live and interact with human users in a very intuitive way. The project aims to contribute

to the advancement of robots’ capabilities in comprehensively understanding human beings and appropriately adapting their behaviour to the context to accommodate unknown and changing environments, tasks and users. The project

is developing two innovations with commercial potential: an adaptive emotional engagement and a probabilistic sensor fusion for personalised social interaction.




23

Similarly, the Donating Robots a Theory of Mind (DoRoThy) project focuses on

the meta-cognitive functions of robots so that they can improve their understanding of human cognition and be able to interpret the social behaviour of

humans. Drawing on the case of child-robot interactions in the frame of education, the project is currently piloting its results in collaboration with schools in the United Kingdom. One of the challenges of such a project is the liability of

robots and use of AI. Although robots are designed to behave in a certain way and be compliant, as raised during the panel discussion there is a possibility of

robot malfunction. In such cases, it is important to clearly identify liability and clarify responsibility in case of a robot’s unexpected behaviour. Beyond unexpected behaviour and liability, this also raises ethical considerations,

especially where children are involved. This regards the responsibility in terms of the ‘deception’ that may be experienced since children (and other social groups)

might have the impression that they are dealing with a fully autonomous machine. Another challenge regards the setting of successful metrics in robot

performance, especially in cases where robots are used for education or therapeutic purposes (i.e. in such cases, performance is not subjective such as in autonomous cars – move from a to b).

The Adaptive Decision support for Agents negotiation in electricity

market and smart grid Power Transactions (ADAPT) project focuses on electricity markets and smart grids. The project aims to support the decision-

making of players in these areas by assisting their advising and negotiation capacity such as that they can optimise the outcomes from the electricity transactions. The project encapsulates a variety of AI approaches to make

recommendations for optimal bids in auction-based energy markets. However, one of the challenges experienced by the project is the lack of access to the data

necessary for testing the approaches suggested by the project on a large scale. In the end, a decision support system for players’ energy negotiations was developed as a product to be commercially exploited. In the case of the ADAPT

project, several parts of the solution are transferred to industry, but not the technology as a whole. This is due to the fact that the main impact of this

technology can be generated when the local markets are open, which is not the case at the present moment. Large companies would be the only partner possible, but they are difficult to engage. However, the project has received

industry-sponsored funding.

The event also included projects of the Digital cluster that were presented in the poster session. These included the Enabling Demand Response for short and

real-time Efficient And Market Based smart Grid Operation - An intelligent and real-time simulation approach (DREAM-GO) that aims to

conceive, develop, implement and validate models enabling Demand Response (DR) for short and real-time efficient and market based smart grid operations. Among others, the project develops models and methods to simulate and assess

the use of short and real-time DR in smart grids.





24

The Hybrid Learning Systems utilizing Sum-Product Networks (HYBSPN)

project combines the complementary advantages of tractable probabilistic models and deep learning to extend previous state-of-the art on probabilistic machine

learning systems.

The Learning and Analysing Massive/Big complex Data (LAMBDA) project develops next generation advanced 3D shape retrieval technology, aiming at

retrieval from a database of hundreds of thousands in the order of seconds. The project methodology includes a mathematically rigorous geometric approach to machine learning and data mining. LAMBDA further implements novel techniques

focusing on improving travel time estimation approaches, identifying different driving behaviour patterns and frequent sub-trajectories of user trajectories.

The poster session also included projects relevant to security and forensics. The

Computer Vision Enabled Multimedia Forensics and People Identification (IDENTITY) project aims to boost the use of biometrics in the forensics field and also promote the integration of multimedia forensics into forensic science.

Along the same lines, the Soft Computing and Computer Vision for Comparative Radiography in Forensic (SKELETON-ID) is developing an automatic forensic identification system based on the comparison of X-ray

images. More specifically, drawing on comparative radiography the project aims to develop a novel AI-based framework that will enable the comparison of images

and will support experts in their decision-making.

The Neuromorphic EMG Processing with Spiking Neural Networks (NEPSpiNN) aims to realise a neuromorphic event-based neural processing system that can directly interface with commercial surface EMG for the extraction

of signal features and classification of the hand-gesture. The signals produced can be used to extract information about desired movements of the subjects under

investigation and, in the context of rehabilitation, to control artificial systems for augmenting the ability of the subject.

Other projects of this cluster presented in the poster session included projects

that tackle the areas of manufacturing and robotics. The Industry 4.0 for SMEs Smart Manufacturing and Logistics for SMEs in an X-to-order and Mass Customization Environment- Erwin (SME 4.0) project focuses on identifying

the requirements for introducing Industry 4.0 in small and medium sized enterprises (SME) and to develop concepts for manufacturing and logistics

systems, as well as adapted organization and business models for smart SME factories. The project explores the use of AI in this respect by analysing the possibilities of AI for automating the design of manufacturing systems, applying

AI for the identification and prediction of operator’s tasks in human-robot collaboration, and developing demonstrators for AI applications in industry for

SMEs.

Moreover, the Towards Intelligent Cognitive AUVs (TIC-AUV) project aims to support the development of smarter Autonomous Underwater Vehicles (AUVs)







https://cordis.europa.eu/project/id/709136/


25

that will have the capacity to remain autonomous for longer time intervals. The

project uses semantic-aided localisation, dynamic real-time planning, together with fault management to enable smarter AUVs that will have the capabilities to

respond to unforeseen events and faults.

Challenges identified by the projects of the Digital cluster

During the panel discussions, the speakers had the opportunity to interact with the audience and to further reflect on their projects and the challenges currently

experienced with the use of AI techniques in their fields. This subsection summarizes the key challenges identified by the projects of the Digital cluster.

#1 Reuse of research results: Undoubtedly, there is high-quality research

conducted by a plethora of projects that make use of AI. Such projects develop cutting-edge AI approaches that could potentially also be used in other domains. Thus, identifying pathways for re-using research results by other researchers or

even by other fields is important. For example, the BROS project highlighted that their approach could also be reused for eliminating corruption incidents. In this

respect, future EC calls should also aim to strengthen and promote open science by encouraging researchers to make the code and datasets publicly available. Following a similar approach to Perkel (2019) described in his Nature article

“Workflow systems turn raw data into scientific knowledge”, the EC could develop workflows or processes to cross-check the reusability of code and datasets used

in projects. Other approaches to encourage the exchange of know-how and reusability of results is to further encourage cross-disciplinary collaboration and research. Following the example of the MIT Media lab, the EC could provide

additional incentives to research centres to team-up researchers from different disciplines within the same research centre or research group.

#2 Connecting with industry: The projects’ significant research results are

often not adequately communicated to industry. Since academic research impact is to a large extent measured in research publications, researchers tend not to

communicate their research results to industry nor to the general public. Beyond publications, impact indicators should be developed in terms of how research results are changing people’s lives e.g., creating a thematic community.

Researchers often mention that their network lacks industry representatives and investors (i.e. business angels, venture capitalists). However, there are multiple

events at international, national and also regional level that bring researchers and investors together. Thus, work has to be done regarding the (online) promotion and widespread dissemination of these events. Along the same lines, researchers

often lack the necessary skills for communicating their research results beyond academia. The next generation of researchers need to be trained not only on how

to present their research results to the academic community, but also on how to present their research in the form of a simplified story. The collaboration with PR officers could also help the researchers’ promotion of their research results and

increase the impact of their research.

https://www.nature.com/articles/d41586-019-02619-z


26

#3 Data handling and privacy: Data handling, privacy, and the security of

sensitive data are among the most frequently highlighted challenges in AI research, especially for projects handling very sensitive personal data. Existing

regulations, such as the GDPR, as well as considerations such as ‘Facial recognition technology: fundamental rights’ raise additional challenges for projects in need of collecting private data. In this respect, ethical

recommendations should guide projects on ensuring that, if collected, appropriate measures for consent, data storage and data security are applied.

#4 Law and Ethics: Researchers agree that there is a need for additional law

enforcement and ethical recommendations on specific aspects of robot use and AI. Reflecting on the objectives of the projects presented in the digital panel, this

could concern aspects such as quality control, deception risk, and legitimacy. Regarding quality control, there is currently a lack of guidelines on how to conduct quality checks of robots before releasing on to the market. Along the

same lines, the risk of deception raises ethical considerations especially for vulnerable groups (e.g., children, people with dementia, autism, etc.) as such

groups may get the impression that they are interacting with a fully autonomous machine and thus deception or impertinence issues might be raised. Building quality assurance approaches and metrics could guide researchers in ensuring the

quality of robot behaviour before allowing interaction with the public. Similarly, guidelines on minimising deception risks are also necessary. Beyond ethical

considerations, there is also a need to clearly address legitimacy issues. Although robots are designed to behave in a certain way, there is always the possibility of misbehaviour. This raises the need for regulation on the legitimacy of robot use.

#5 Education and AI tools: The use of AI in education should certainly not be viewed, somewhat naively, as “machines and programmes replacing teachers”,

but rather as an extra aid, a set of extra tools. AI can play an important role when addressing certain learning difficulties such as dyslexia or lack of concentration. It can provide better personalised help or automated tests,

differentiating between the knowledge level of individual students and the progress in their studies. It might also be used to reduce the still increasing

amount of administrative tasks of teachers at every level of education. In the future, several types of practitioners (e.g., teachers as highlighted in the DoRoThy project) are expected to make use of robots and AI. Thus, there is a

need to proactively train the different professions on the benefits and risks of using these technologies. Moreover, since children and other groups of the

population will use these technologies, there is a further challenge on how to encourage the use of such technologies while avoiding addiction. One solution

could be to include courses that will train children on machine-learning to understand what AI and robots are and their functionality. This might help them understand that such machines have limitations and that they cannot replace

human-to-human interaction.

#6 Extending innovation potential: There is a lack of support at the moment to create a minimal viable product. As research results can be commercialised, it

https://fra.europa.eu/en/publication/2019/facial-recognition

https://fra.europa.eu/en/publication/2019/facial-recognition


27

is necessary for researchers to acquire the necessary resources needed for spin-

out company creation, reliability testing, software licensing, etc. However, the majority of the researchers lack the necessary business skills. As the fellow of the

DoRoThy project rightly highlighted, researchers often do not have the skill-sets necessary to develop a product or a business, or to write a business plan or to pitch to an investor. There are few entrepreneurial researchers able to

understand both the scientific and the business worlds. In this respect, inter-sectoral cooperation and the active involvement of private companies as

beneficiaries in projects could contribute to the development of the researchers’ skills by coaching and mentoring them in business matters.

#7 Training the researchers: AI is a broad multidisciplinary area applicable in

a plethora of fields. Beyond computer scientists, researchers in other fields (e.g., engineers, social scientists, economists etc.) often lack training on AI. The majority of such researchers interested in AI draw on “self-training” in order to

acquire the necessary knowledge. However, to be able to unlock the full potential of AI’s applicability in the different fields it is important that research centres

provide training on AI to their researchers. Such training curricula could take the form of other types of horizontal training provided in research centres such as research methods, presentation skills, etc.

3.2 AI in MSCA HEALTH projects

Overview of the project presentations

Aiming to assist healthcare practitioners in their clinical decisions, the Rapid Biomechanics Simulation for Personalized Clinical Design (RAINBOW)

project draws on computational medicine and ICT. The project focuses on providing patient-specific simulation models that can help clinical experts in their decisions without the assistance of technical experts.

Similarly, the A pEdiatRic dosimetRy personalized platfORm based on computational anthropomorphic phantoms (ERROR) project also aims to assist the decision-making of clinicians in their diagnostic and therapeutic

protocols. In particular, ERROR focuses on assisting the personalisation of dosimetry in paediatric clinical protocols where there is currently a lack of

standardised dosimetry protocols. The project employs simulation techniques and computational models for the implementation of its software. One of the challenges highlighted by both the RAINBOW and ERROR projects is associated

with accessing data for infusing their proposed approaches. In particular, one of the researchers noted that patient data needs to remain in the hospital and

cannot be exported, and that this is limiting the capacity of their projects to access data. Beyond this, panellists also raised questions on whether decision-making assistance tools being developed can be considered as fully automated

and reliable. Given that such tools are to be employed in the healthcare sector, it is important to determine the extent to which they can be accurate. As agreed by

all panellists, such tools are supplementary to the doctor’s own perception, experience and decision-making.




28

The Automated Measurement of Engagement Level of Children with

Autism Spectrum Conditions during Human-robot Interaction (EngageME) project assists the educational therapy of children with ASC (Autism Spectrum

Conditions) through their participation in communication-centred activities. The project employs humanoid robots and novel machine learning models to effectively leverage multimodal behavioural cues. The project holds the potential

to make a significant contribution in this field. However, as noted during the panel discussion, one of the major challenges relevant to the potential impact of

this project is relevant to the cost associated with purchasing the necessary technology.

The A Biomimetic Learning Control Scheme for control of Modular Robots

(BIOMODULAR) project focuses on the development of artificial adaptive learning systems to develop a modular biomimetic architecture for controlling robots, and to get insights into the brain capability to achieve motor adaptation and learning.

The project’s approach is based on the idea of mimicking the biological functionality of the central nervous system, which will lead to the creation of

autonomous intelligent robots and thus aims at assisting robots to operate in dynamically changing environments. During the panel discussion, questions were raised regarding the subsequent ethical and legal implications of this type of

research. Ethical challenges are relevant to the extent to which the creation of autonomous intelligent robots that mimic human brain behaviour will lead to the

replacement of human-to-human interaction with human-to-robot interaction. Additionally, legal challenges arise, relevant to robot malfunctioning and control.

Reflecting on the need to identify malignant breast tumours in real-time, the

Microwave Diagnosis of Breast Cancer with Open Ended Contact Probes (MIDxPRO) project aims to contribute to the diagnosis and treatment of breast cancer. In particular, the project proposes a microwave technology to diagnose

suspicious samples simultaneously and autonomously such that accurate and fast assessment of samples can be achieved. The project employs supervised machine

learning and feature selection techniques to collect data to improve the accuracy of the method.

Similarly, the Training Network on Automatic Processing of PAthological Speech (TAPAS) project is working on assisting the diagnosis, treatment and

well-being of individuals with spoken communication difficulties (e.g., impairments in speech production, speech perception, language processing). In

particular, the project contributes to the development of tools for the detection, assessment and treatment of pathological speech as well as of technologies for the assisted living and care of individuals with speech pathologies. As raised

during the panel discussion, many of the projects employ AI and data analytics tools to assist health practitioners in their diagnosis and treatment effort.

However, such methods do not guarantee accuracy and thus, healthcare practitioners may be sceptical about their adoption. One of the challenges in the healthcare field is the interpretability of the suggestions of these tools by the

doctors, nurses and carers.






29

The Outcomes of Patients’ Evidence With Novel, Do-It-Yourself Artificial

Pancreas Technology (OPEN) project aims to assist people with diabetes and aims to establish a more robust evidence-base knowledge on the impact of do-it-

yourself artificial pancreas systems. The project will extract insights on the impact of the clinical condition and quality of life of the individuals so as to inform technological improvements (e.g., the predictive capacity of such systems) and

generate valuable conclusions that will accelerate improvements and diffusion of APS technology across the wider population of people living with diabetes.

Similarly, Deep Understanding of Patient Experience of Healthcare from

Social Media (DeepPatient) develops an automated tool to process large-scale social media so as to assist the in-depth understanding of patient experience.

This is then processed and mapped into various aspects of healthcare services to discover connections between elements that result in a perception of low and high quality of service. To achieve these goals, the project has developed a

neural topic model for aspect extraction (e.g., Bayesian models and dimensionality reduction), a novel framework based on reinforcement learning for

topic modelling in order to improve the topic coherence measures, and has suggested a joint learning method with Generative Adversarial Network to improve the performance both in aspect extraction and sentiment classification.

Both OPEN and DeepPatient projects draw on sensitive data of individuals to derive conclusions. One of the challenges that needs to be addressed in this type

of research is how to achieve consent for any personal data that is used, even if this data is extracted from social networks or the web.

The poster session also included projects relevant to the health cluster. In

particular, the Safe, Efficient and Integrated Indoor Robotic Fleet for Logistic Applications in Healthcare and Commercial Spaces (ENDORSE) project aims to develop a safe, efficient and integrated indoor robotic fleet for

logistic applications aimed at providing an autonomous eHealth robotic module that will facilitate electronic health records connectivity. The project draws on

machine learning techniques for developing autonomous navigation and advanced human-robot interaction solutions aiming to leverage deep learning algorithms for robotic vision, and act as a data source in the broader digital healthcare context,

enabling artificial intelligence solutions for personalizing care.

The Physiological and Rehabilitation Outcomes: Gains from Automated Interventions in stroke Therapy (PRO GAIT) focuses on developing an

intelligent robotic gait training device that can interpret and respond to user intent after a stroke. The project aims to capture and model/exploit EMG and EEG biosignals during robotic gait after a stroke. This includes the development of a

large repository of co-registered biosignals during robotic walking in healthy individuals and those after a stroke, and the exploration of advanced algorithm

development to decode movement from EEG/EMG data and applying findings to early prototype development.






30

Challenges identified by the projects of the Health cluster

#1 Access to data: One of the key challenges of the projects in this cluster is access to data. Due to the sensitivity of the data needed, even if available, hospitals and other healthcare organisations cannot fulfil the data needs of such

research projects. One pathway to address this issue is to encourage healthcare organisations to formulate specific teams within their organisations which will be

in charge of assisting such requests from researchers. In such cases, researchers could provide their algorithms/tools/data-mining techniques to the team to run it internally such that patient data can remain internal to the healthcare

organisations and direct access to the datasets by the researchers can be prevented. Similar approaches are already followed by some hospitals in the U.K.

EU guidelines could be put in place to support and guide hospitals and other relevant organisations on how they manage such departments internally, advising them on data-handling issues in order to protect patients and at the same time

be able to support researchers.

#2 Data anonymisation: Another challenge is related to difficulties encountered in data anonymisation. Even if researchers are able to gain data access, making

sure that the data is fully anonymized is not always possible. Some of the projects in the healthcare sector involve profiling individuals including children

(e.g., EngageME). In certain types of projects, there is no one-size-fits-all solution that can be applicable to every individual case. For the case of autism, it is necessary to collect data for each different individual, as every individual is

different and therefore anonymisation is impossible. However, this raises ethical issues for the handling and processing of such data.

#3 Skills development and engagement of the practitioner: Some of the

projects presented aim to provide solutions that will support clinicians and will have an advisory/supporting role. Convincing and training practitioners in this respect is a challenge for the healthcare domain. In the future, it is more likely

that such automated introversion will be widely adopted. Certainly, the doctor will remain in the loop, but such tools will contribute to the reduction of significant

costs in the healthcare sector and is likely to pose notable changes on the focus of the future profession. Thus, assisting healthcare practitioners in understanding

and using AI tools is important. Beyond training, it is necessary to assist practitioners to understand the functionality of AI such that wider acceptance and engagement in the use of such tools is possible. The fact that AI is not 100%

accurate might pose risks if misused, especially in the healthcare sector. However, the interpretability of the suggestions provided by AI tools could

potentially assist practitioners to achieve better decision-making and improve the accuracy in their diagnoses and approaches.

#4 Cost: AI tools and techniques are expected to improve the quality of life of patients and can generate significant savings for the healthcare system in the

long term. However, one challenge is the upfront cost of the handler to deploy this kind of technology hospitals (e.g. 40.000€ for a robot is very expensive). A


31

potential solution to this problem is to encourage projects to also provide

alternatives to the use of expensive technology such as the development of software that is also functional with simpler devices e.g. camera, microphone, or

wearable.

3.3 AI in MSCA ENVIRONMENT projects

Overview of project presentations

The Deep Learning UAV Networks for Autonomous Forest Firefighting (DUF) project tackles forest fire fighting through the development of novel multi-

agent deep reinforcement learning algorithms for enabling the coordination of multiple autonomous drones working towards forest fire fighting. Extracting

information through the utilisation of Unmanned Aerial Vehicles (UAVs), the project uses multi-agent deep learning techniques to provide estimates of the fire

spread direction. It also provides a novel graphical encoding based on a deep reinforcement-learning algorithm which is utilised to coordinate UAVs actions for suppressing the fire.

The focus of the Probabilistic neuromorphic architecture for real-time

Sensor fusion applied to Smart, water quality monitoring systems (ProbSenS) project is the monitoring of water pollutants through electrochemical

analysis of fluids in real time. Inspired by how our brain works, the architecture developed in the project employs neuromorphic networks to fuse multivariate microsensor data in order to efficiently monitor water quality in situ and in real

time, and to do it resiliently to sensor non-idealities. The ProbSenS project is particularly innovative, since it filed a patent application with two organisations as

applicants. The researchers are collaborating with a company in the creation of a database (another type of intellectual property right). However, the speaker noted that sharing information from the database, which is not related to the

content of the patent application, had to be balanced.

The Internet of Things for Smart Water Innovative Networks (IoT4Win) project focuses on the management and transport of finite water resources

through the use of advanced sensors and IoT technology for water and environment monitoring and control. In particular, the project draws on smart

sensing and communication within energy limited heterogeneous devices in IoT enabled urban water environment; dynamic sensor web and interoperable open platform with Integrated Knowledge Management for smart water networks; data

security and intelligence in IoT enabled smart water network.

Finally, the Low Energy Dew Point Cooling for Computing Data Centres (DEW) project focuses on the cooling of computing data centres (CDCs) aiming to

reduce electricity consumption, currently estimated to be 1.3% of total world energy consumption. The project aims to develop a computerised tool and technology prototypes for a novel computer and data centre dew point cooling

system. The project is expected to achieve electrical energy savings of up to 60% to 90% in CDCs and thus decrease the initial costs compared to the traditional






32

CDC air conditioning systems, which will result in removing the outstanding

existing problems with the CDC cooling systems. One of the major challenges of such projects is access to data. As noted by one of the researchers, regulations in

some countries prevent researchers from gaining access to the real data necessary for the implementation of their approaches. Adding to this discussion, another researcher highlighted the urgent need of pilot sides that can offer large

scale experimentation. The researchers pointed out that AI has been good in resolving small problems, where simulations can also be employed, but it is

important to be able to experiment in the real world where large-scale energy systems exist. However, in most of the cases, this is impossible and hence reliability issues arising by current research restrict wide use of AI in the energy

sector.

In addition to the projects that were presented, the poster session also included projects relevant to the environment cluster. For example, the Bio-inspired

Technologies for a Sustainable Marine Ecosystem (ECOBOTICS.SEA) employs AI techniques which are applied mainly for the analysis and

characterisation of the species in the wild, by modelling using probabilistic approaches and novel computing architectures.

The focus of the Sustainable energy demand side management for GREEN Data Centres (GREENDC) project is energy efficiency in data centres achieved

through the use of a decision support tool that helps data centre managers to better predict energy demands and evaluate operation strategies to minimise

energy waste, cost and minimise CO2 emissions indirectly. In particular, the project adopts a non-linear energy forecasting model and provides a simulation

tool to allow data centre managers to conduct what-if analyses on the number of cooling devices to be turned on. This also enables detecting the target temperatures of each cooling device and controling the number of virtual

machines to minimise heat generation by servers, among others considering factors for energy demands and supply.

The SENSors and Intelligence in BuLt Environment (SENSIBLE) project

focuses on developing technology solutions to ensure smart buildings deliver on their energy efficiency and sustainability targets while accounting for the occupant’s perception of comfort and well-being. The project encapsulates an

integrated research and innovation approach that includes sensor development, adaptive connectivity enabling real-time condition monitoring and AI for providing

actionable inference.

The Holistic Surface Water and Groundwater Management for Sustainable Cities (Water4Cities) project draws on sensor technologies, improved data collection,

analytics and new visualisation tools to enable the smart localisation, visualisation and analysis of urban water (both surface water and groundwater) at a holistic urban setting to assist the real-time monitoring of water resources and provide

support on decisions for optimal urban water management. As commercially






33

exploitable outcomes, the project is also developing an open platform with

decision support services for holistic surface and groundwater management.

The poster session also included cross-disciplinary projects in the fields of environment and agriculture. The Accounting for Climate Change in Water

and Agriculture management (ACCWA) project aims to develop remote-sensing based management and monitoring tools for food security and water and

agricultural risk management that allow improving the reliability of decision-making regarding water use, yield and hazards in agriculture. The project employs AI to derive high-resolution soil moisture at the field scale, to map

irrigated areas, for yield forecast over dryland areas, and for the monitoring of water resources at high resolution from gravity data.

Along the same lines, the Development of an Easy-to-use Metagenomics

Platform for Agricultural Science (MetaPlat) project aims at creating an easy-to-use integrated hardware and software platform to enable the rapid analysis of large metagenomic datasets, providing actionable insights into probiotic

supplement usage, methane production and feed conversion efficiency in cattle. The project is developing faster and more accurate analytic platforms in order to

fully utilise the datasets generated to study the change within microbial communities, under various conditions in cattle guts and impacting probiotic supplement usage, methane production and feed conversion efficiency.

The Enhancing and implementing Knowledge based ICT solutions within high Risk and Uncertain Conditions for Agriculture Production Systems (RUC-APS) aims to develop an innovative technology-based knowledge

management system to capture agricultural information in terms of collecting, storing, processing, and disseminating information about uncertain environmental

conditions that affect agricultural decision-making production systems. As such, the project planned to provide knowledge on every stage of the agricultural-cycle so as to assist the realisation of the key impacts of every stage of agriculture-

related processes.

Challenges identified by the projects of the Environment cluster

This cluster included one panel discussion during which all project presenters were invited to share their thoughts on the challenges arising from the use of AI

for the environment. This section summarises the key outcomes of the panel discussion.

#1 Data access: Most of the researchers in the field cannot gain access to data,

or find that data are scarce or expensive to obtain. As a result, many advanced AI techniques cannot be used. For example, as highlighted by the DEW project,

access to data on cooling systems in data centres is in some countries (e.g., U.K.) even prohibited by law. Lack of data for the training of the AI algorithms is the main constraint. For this reason, the majority of the projects are selecting

alternative approaches such as training AI algorithms. However, there is a need to enable researchers to access data and test their approaches in the real world.





34

The EU should encourage organisations to collaborate with researchers as part of

their corporate social responsibility, and provide access to infrastructure for experimentation, etc. Changing the culture of data-sharing, where possible,

would be tremendously positive to create open-access and larger databases.

#2 Interdisciplinary research: Scientists in this field lack the necessary skills to detect which technologies are suitable for resolving environmental problems.

Facilitating interdisciplinary research and fostering dialogue between researchers in computer science and environment raises the potential for the use of cutting-edge AI techniques in the environment field. Additional incentives that will

encourage researchers in this field to experiment using emerging technologies would also be helpful.

#3 Engagement of citizens: As noted during the panel discussion, despite the

techniques developed by scientists in the field, one of the challenges is how to convince the wider public from different specialisations to not only engage with the suggested tools and approaches, but also to innovate for the environment.

The further engagement of citizens, especially younger citizens, in environmental matters is of core importance. Thus, deriving additional approaches for promoting

awareness and engagement is necessary. This might include the creation of citizens’ labs to encourage citizens to develop AI tools and software for environmental issues or the use.

#4 Supporting proof of concept: Many projects develop approaches that have the potential to be widely applied and exploited in the market. However, it is often difficult to gain proof-of-concept as the nature of the projects might require

testing on a larger scale. Thus, the infrastructure necessary for gaining proof-of-concept could be impossible, or too expensive to build. The European Commission

could support projects or clusters of projects in building common infrastructure that will allow them to run simulations and tests such that they can gain proof-of-concept.

3.4 Cross-cutting challenges for AI research

The three panels addressed a series of challenges which are cross-cutting for all fields of AI applications. Below is a short summary of these main challenges:

Good and reliable communication on the limits of AI research methods and

applications. AI is often perceived as being the “silver bullet” for everything and sometimes is “oversold.” The limits of AI methods must be communicated and understood. AI only works well when reliable structured

data is available. However, when all required data is not available, other techniques like “old fashioned modelling” are preferable.

Responsible AI research. AI must balance the needs for being faster, more accurate and more efficient with avoiding biases; AI must be accountable and inclusive.

Synergies and opportunities in AI across fields. AI methods need to be explainable at many levels: to the end-user; at the engineer/developer


35

level, and at the regulator levels for ethics and legal issues, to name a few.

In the health sector, for instance, the goal of such integration of methods could be the development of tools to link biology, medical physics (image

processing), radiobiology for diagnosing and treating various diseases. Data collection and processing. A key challenge in all fields is to ensure

that enough data is collected which is of high enough quality. In most

cases, the collection itself can be cumbersome (e.g. applying new diagnostic technologies in health). In the environmental field, a main

challenge is having data from real system operations. Nonetheless, how to ensure the privacy of the data used and how to collect large-scale labelled training data for the deep learning models?

Safety and security. In the field of robotics, the Internet of Things and cyber-physical systems in general, ensuring safety and trust are key. This

is especially relevant when an autonomous system is controlled using a learning based-method and requires “safe learning”/”secure learning”

guarantees. From these challenges, the key needs for policy consideration are grouped into six key areas, as elaborated below, with practical examples retrieved from the

panel discussions.

#1 Interdisciplinarity

Although the event highlighted that a plethora of MCSA projects from different backgrounds conduct high-quality research on AI, there is an emerging need to further expand the interdisciplinarity of research projects. First, different skill-sets

are required to ensure that the full potential of AI in each field will be exploited. Currently, as highlighted by the event participants, there is still a limited

exchange of know-how between researchers of different disciplines. Thus, there is an urgent need to further promote the collaboration and exchange of know-how. Along the same lines, there is a need for more scientists to work in

interdisciplinary fields to close the gap (e.g. between software developers and clinicians, social scientists and computer scientists). Second, despite the high-

quality of the research conducted on AI, there is currently low motivation and capability of exploring the potential re-usability of the results by researchers in

other fields. As clarified by some of the speakers, the AI techniques and approaches developed may be potentially applicable for other researchers or even by other fields. It is necessary that additional pathways are identified for assisting

AI researchers to communicate their research results to other disciplines and promote the reusability of AI techniques in different domains.

Linked to the idea of interdisciplinary is the consideration of “Hybrid AI”. AI is capable of providing answers, supporting decision-making or even automating tasks in several domains, including health and education. This poses a potential

for enhancing efficiency and reducing costs. However, there is a natural fear of AI, mainly because the concept itself is not well understood, or even strongly misunderstood. Although some tasks will achieve greater automation because of


36

AI, it should be communicated that this is expected to be a smooth and slow

transformation in which humans will remain in the loop and certainly in control. This will ensure that final decisions are controlled by decent security and ethical

mechanisms. At the same time, the role of practitioners in several fields will be transformed, enabling them to save a lot of time and effort to focus on challenging tasks and optimally target the most challenging problems. For

example, some of the projects presented aim to provide AI solutions that will support clinicians in their diagnostic approaches. These will have an

advisory/supporting role. As highlighted during the meeting, this will enable practitioners to focus more on difficult situations.

#2 Training and skills needed in the field

Different groups of the population, including researchers, practitioners, policy

makers, research funders and the general public will require a level of training on AI.

With the exception of computer scientists, researchers currently lack a basic

understanding of AI’s capabilities and limitations that would help them assess its applicability in their field. Researchers from different fields will need to be trained

on AI techniques and on how to acquire data science skills, so that interdisciplinary applications are realised. Research centres could possibly formulate course curricular for providing training on AI.

Practitioner training is needed to understand and learn the basic concepts of AI,

the system development behind AI, and finally how to use AI. This will assist practitioners in understanding the functionality of such tools so as to understand

the extent of their reliability and their limitations. They should have access to training modules on AI, including general concept of mathematical modelling, optimisation, probability and statistics, algorithmics and programming basics.

This should happen already at the Bachelor level (latest Master level).

Policy-makers and research funders need to encourage a broader knowledge of AI to ensure support for a new generation of researchers. The dependency on single

experts brings a lot of risks in research projects. Awareness about possibilities of developments and applications in all fields is necessary. They should also foster

interactions between different consortia, projects and programmes by giving extra funding dedicated to synergy building and/or ensuring methods and data are open sourced.

The general public should be trained in order to get a grasp on the “what and the

why of AI” and be better informed on its uses and its abuses There is a natural fear of AI, as already said previously in this report, mainly because the concept

itself is not well-understood. People use smartphones and similar tools on a daily basis, as, for example, any virtual assistant when asking about the weather or

the next train, spelling and grammar checkers; they are thus using AI tools almost on a permanent basis. Still, AI is broadly viewed as a threat. As we are moving towards a future of personalised, widespread robots assistance, we need


37

to ensure that the capabilities and the limitations of these technologies are well

understood.

Knowledge and training in trustworthiness and ethics of AI covering research design, impacts, and consequences from applications is key for all groups. AI is

not just a technical science (i.e. coupling computer scientists with biomedical engineers) and it is crucial that it gets the needed social science inputs into its

developments and field applications. AI is not “magic” and will have greater or less success depending on the specific cases of application. Last but not least, the final decisions should depend on human experts and the AI systems will not

replace the human element, but will instead support or complement them. In other words, no matter how intelligent a machine is, it does not solve problems,

but instead executes solutions.

#3 Governance (ethics and regulation)

The “Ethics guidelines for trustworthy AI” certainly provide guidance to the ecosystem on how to use and to operationalise AI in a way that they find useful.

Ensuring ethical use of AI in terms of accountability of outcomes, bias-free outcomes and removing any harmful implications of AI in research projects is key. Another ethics risk is rooted in the observation that whereas human beings

have moral capacities, algorithms do not. Human empathy leads us frequently to “forgive,” to allow exceptions in particular situations, sometimes in a somewhat

unpredictable way. Algorithms do not feel pain or regret and are incapable of this kind of behaviour. Machines miss social conscience. In this light, it is extremely important to keep control and to always view results of algorithms that are used

in decision-making in their proper context, i.e., they are tools that only give unemotional advice based on cold, technical facts. We should not always expect

useful input in this from developers, from engineers. Their answer is frequently that they just develop the technology, that they do not determine how it is used. Policy-makers have a responsibility to make the necessary corrections here. In

addition, practical examples of how to use ethical guidelines to ensure the security, privacy, and integrity of project results could be helpful. This could also

potentially take the form of a toolbox that could assist projects in the ‘certification’ of pilot studies with respect to ethical matters. Nevertheless as

highlighted by the audience, ethics guidelines are different to regulation, which might be possible and useful in some cases. Thus, the further regulation ‘of what could be potentially regulated’ on AI needs to be considered such that the

appropriate and necessary legal framework can be constructed. Topics on AI that could be potentially regulated may be relevant to the liability of AI or robot use.

#4 Access to data

The use of AI techniques requires access to large pools of insightful data.

However, in most cases, data are scarce, expensive to obtain or data access is restricted due to privacy issues. This often prevents researchers from employing

AI or from unlocking its full potential. The development of incentives for



38

organisations (e.g., healthcare organisations, private companies) for data

anonymization, where possible, or assisting the researchers to run their algorithms internally would be desirable. With the recent new European directive,

the regulatory framework for sharing data is more stringent. An adequate Information Governance structure is essential in public organisations and training for staff in this subject is becoming mandatory every year. One of the serious

risks related to the use of AI is that it aims (and claims) to be objective, but it frequently perpetuates mistakes and prejudices of the past, since the underlying

machine learning is usually mainly trained by examples based on decisions from this same past. For example, if a company was prone in the past to hire men rather than women, this discrimination is likely to be continued by an AI system if

it is fed with data reflecting this. Let us consider, as another example, the training of autonomous cars. The question raised is how should the car react if

some children and some elderly people cross the street at the same time and only one group may be avoided. Training material (human answers) will most

certainly provide quite different answers in Japan and Europe, in view of their cultural differences. Having more data is not necessarily a guarantee that the information obtained from these is more precise, more relevant, up-to-date or

even really useful. In particular, existing correlations in the past (“being a woman” and “being underpaid”) should certainly not survive in the future.

#5 Innovation management

Innovation management is relevant to AI, in particular in view of the extreme

change in paradigm and the risk of the presence of non-transparent “black box” algorithms developed by intelligent systems. There is a strong need for new

proof-of-concept instruments and extended criteria to evaluate innovation potential, in particular when evaluating AI related MSCA proposals. Emphasis should also be put on the re-use and re-usability of previously obtained research

results. In addition, intellectual property rights in AI consist mainly of copyright (i.e. computer language), databases (collection of data) and, although more

challenging to register in this technology context, patent applications. Other types of rights such as know-how - also known as trade secrets - and trademarks are key intangible assets and should also be considered. Thus, intellectual property

rights should be carefully assessed before exploitation, whether commercial or not, including dissemination.

The development of innovations requires funding and resources, which can also

originate from private entities. When engaging with companies or investors, researchers rarely have the required set of skills to negotiate a win-win deal.

Therefore, public organisations nowadays have a Technology Transfer Office or Innovation Department (or similar name), composed of individuals with scientific backgrounds, legal knowledge and business experience, who are able to support

and guide academics through the process of interacting with the private sector.


39

#6 Realistic communication and quality assurance

There is a risk that AI can be a concept that is “oversold” and, through effective communication, this should be prevented. AI should not be regarded as the silver bullet for all problems. AI is only capable of producing the desired outcomes if a

good amount of structured data is available. If that is not the case, then traditional modelling is likely to be more appropriate. The capabilities and

limitations of AI should be well understood by the researchers so that its use can be appropriately justified.

There is a natural fear of the unknown. AI is not well understood or

misunderstood. It is important to create ways of engaging and communicating with the public about the uses/misuses and potential future of AI. The public must feel AI is trustworthy, appropriately used and ethically sound.

4. Meeting conclusions

The event managed to successfully achieve its planned objectives, based on the

feedback received both during the meeting and after, in the survey shared with all the participants. Thus, the 37 invited MSCA projects showcased their ongoing or completed work and highlighted their contribution to European AI research in

the digital, health and environment sectors. By bringing together these MSCA success stories in AI research, new contacts and collaborations between scientists

have been fostered and new ideas in their scientific disciplines have been generated. Moreover, the interaction between research practitioners and policy-makers contributed to raising awareness of the challenges of the AI research and

also to outlining future policy needs.

The participants appreciated the clear policy framework and the presentations by the European Commission representatives, who raised awareness of the

consultation processes, the areas where scientists could give inputs, or follow-up pilot initiatives where feedback is needed to develop and uptake policies. This is

undoubtedly good practice for co-creation, or joint policy making. Moreover, the support of the experts, not only during the meeting, but also in the preparation of this report, has been very much appreciated.

AI is expected to go through substantial changes, yet without necessarily being

radical and immediate. For the field of health, there was not a fear that AI would take over all jobs in the next 3-5 years, but certainly new skills and training will

be needed to allow effective decision-making in diagnosis and therapy. For technology innovation (e.g. DIY pancreas), these expectations are to become reality. From the digital world, with the growing availability of data from sensor

systems, including 5G and smart grid, there is an increased trend to move away from model-driven analytics to data-driven analytics enabled by AI to provide

richer, more relevant and more meaningful inference.

Regardless of the sector, the participants acknowledged the potential of AI applications to ensure “time saving”. This will allow different concerned


40

employees, whether doctors, teachers, or firefighters, to use their skills for more

important (strategic or technical) issues and/or be focussed more on the human elements of their work. In research, AI could make things much easier and

accelerate turnaround times for certain types of problems, also in terms of the time taken between technology development and real-life useful application developments.

In terms of innovation potential, MSCA projects with strong industrial participation are ready to go further and some are already patenting the main outcomes and are co-creating companies to transfer the research outcomes to

society. In this area, private companies have a leading role in the transfer of knowledge and technology, from the initial idea to the development of a product.

The Innovation Radar initiative was also cited as a useful tool to understand the innovation potential of the projects and to learn about licencing algorithms.

When asked about the focus of the European Commission’s future AI policies, in the feedback survey the participants highlighted the following interesting

suggestions: - Humanity, social responsibility and sustainability: integration of AI

processes into everyday life (e.g. clinical settings); AI should be applied

to societal challenges and sustainability goals; - Innovation: ensure policies that spread EU funds into all fields of

relevance, not just in what seems to be the most “trendy”, or challenging science (e.g. robotics/human-like robots); support innovation for quick and easy adaptation of work processes to AI

products; - Ethics and security: ethical aspects should be in the foreground of any

AI policy development, especially in terms of accountability of outcomes, bias-free outcomes and removal, or avoidance of harmful outcomes and misuse;

- Openness and synergies: ensuring AI computing resources are not only hooked up on large industrial companies, but that free, open

alternatives exist (e.g. open platforms for using AI); making tools and data available in various R&I communities; facilitating interconnection of projects using AI applications;

- Awareness - communicating effectively about AI representing a wide family of techniques that can be very useful for many practical

problems.

These policy areas match well those identified by the expert group in the report.

There is clearly more than technology to consider and the human dimension and ethics are priorities for EU AI policy.

The MSCA researchers and their project outcomes can contribute to policy

development through the joint policy making mechanism. They provide valuable background to understanding developments and trends in different fields. The connection between operational researchers, their supervisors and/or

https://www.innoradar.eu/


41

coordinators and policy counterparts is deemed essential. Researchers also

consider it very important that their skill development and career prospects are acknowledged as providing Excellent Science for scientific support to policy.

For the Policy Officers it is important to be in touch with the actual subject matter

and outcomes of science funded by EU. As a bottom-up programme, the MSCA can complement policy-making in all R&I areas. For this, a careful mapping of the

MSCA project portfolio is central to harnessing the programme’s potential. The creation of synergies with the different European Commission services is crucial in order to maximise impact with all the resources available. Taking stock of the

MSCA projects and their contribution to various EU policy areas can play an important role in achieving efficient joint policy-making.

As the MSCA is also about researcher careers, the challenges faced by

researchers and research organisations (academic and non-academic) should also be taken into consideration for policy-making.

This meeting was a pilot and, as such, it gathered a lot of suggestions for future

similar events (e.g. format and topics in line with the Horizon Europe missions, but not only). MSCA projects have important clustering potential and researchers are very much interested in showcasing their contribution to EU policies and

highlighting the scientific, technological and also societal impact of their work

Fostering collaborations with the different sectors of the European Commission where AI affects policymaking and research priorities is one of the most

important outcomes of this MSCA AI cluster meeting. We are very thankful for the dedicated participation of the researchers who joined the meeting and are confindent that the interactions and networking between scientists, policy-makers

and administrators of the research porfolios has been beneficial for all parties concerned and can only create further synergies for Europe and beyond.


42

References

Marie Sklodowska-Curie Actions Artificial Intelligence Cluster Meeting (event webpage)

Marie Skłodowska Curie actions - Facts & Figures (2019) Factsheet Marie Skłodowska-Curie Actions - Driving innovation, cultivating

excellence in doctoral and postdoctoral training (2019) European Commission (2018). Artificial Intelligence for Europe, COM/2018/237

final. European Commission, Joint Research Centre (2018). Artificial Intelligence: a

European perspective. European Commission, Joint Research Centre (2018). The Impact of Artificial

Intelligence on Learning, Teaching, and Education. European Commission (2019). Factsheet: Artificial Intelligence for Europe. European Commission’s High-Level Expert Group on Artificial Intelligence (2019).

A definition of Artificial Intelligence: main capabilities and scientific disciplines. European Commission’s High-Level Expert Group on Artificial Intelligence (2019).

Ethics guidelines for trustworthy AI. European Commission’s High-Level Expert Group on Artificial Intelligence (2019) Policy and investment recommendations for trustworthy Artificial Intelligence.

European Commission (2019). Trustworthy AI - Brochure. European Commission (2020). Factsheets: a Europe fit for the Digital Age.

European Commission (2020). White Paper on Artificial Intelligence: a European approach to excellence and trust. European Commission (2020). Excellence and trust in Artificial Intelligence.

European Commission (2020). The European Data Strategy. European Commission (2020). Commission Report on safety and liability

implications of AI, the Internet of Things and Robotics. European Commission (2020). Communication: Shaping Europe’s digital future. European Union Agency for Fundamental rights (2019). Facial recognition

technology: fundamental rights considerations in the context of law enforcement. Political Guidelines for the next European Commission 2019-2024.

Perkel, Jeffrey M. (2019). Workflow systems turn raw data into scientific knowledge, Nature 573, 149-150 (2019).

https://ec.europa.eu/info/horizon-2020-msca-cluster-event-artificial-intelligence-10-11-december-2019-brussels_en




https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=COM%3A2018%3A237%3AFIN



https://ec.europa.eu/jrc/en/publication/impact-artificial-intelligence-learning-teaching-and-education

https://ec.europa.eu/jrc/en/publication/impact-artificial-intelligence-learning-teaching-and-education

https://ec.europa.eu/digital-single-market/en/news/factsheet-artificial-intelligence-europe



https://ec.europa.eu/digital-single-market/en/news/policy-and-investment-recommendations-trustworthy-artificial-intelligence

https://ec.europa.eu/digital-single-market/en/news/trustworthy-ai-brochure

https://ec.europa.eu/info/publications/factsheets-europe-fit-digital-age_en

https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en

https://ec.europa.eu/info/publications/white-paper-artificial-intelligence-european-approach-excellence-and-trust_en

https://ec.europa.eu/commission/presscorner/detail/en/FS_20_282

https://ec.europa.eu/commission/presscorner/detail/en/FS_20_283

https://ec.europa.eu/info/publications/commission-report-safety-and-liability-implications-ai-internet-things-and-robotics-0_en

https://ec.europa.eu/info/publications/commission-report-safety-and-liability-implications-ai-internet-things-and-robotics-0_en

https://ec.europa.eu/info/publications/communication-shaping-europes-digital-future_en

https://fra.europa.eu/en/publication/2019/facial-recognition-technology-fundamental-rights-considerations-context-law

https://fra.europa.eu/en/publication/2019/facial-recognition-technology-fundamental-rights-considerations-context-law

https://ec.europa.eu/commission/sites/beta-political/files/political-guidelines-next-commission_en.pdf




43

Annexes

Annex 1: Meeting Agenda

Marie Skłodowska-Curie Actions - Artificial Intelligence Cluster

10-11 December 2019

Research Executive Agency, Covent Garden, Place Rogier 16, B-1210, Brussels, Belgium

DAY 1 – Tuesday 10 December (COVE A2 4th floor auditorium)

13:00-13:30 Participants registration and poster area set-up

Opening session

Artificial Intelligence in European Union and Marie Skłodowska-Curie Actions Chair Fredrik OLSSON HECTOR (REA)

13:30-13:45 Artificial Intelligence in Marie Skłodowska-Curie Actions: Claire MOREL (DG EAC), Alessandra LUCHETTI (REA)

13:45-14:00 Artificial Intelligence strategy for Europe, Cécile HUET (DG CNECT) 14:00-14:30 Keynote presentation: De-mystifying Artificial Intelligence, Alain VERSCHOREN (University of Antwerp)

Panel 1 Artificial Intelligence in DIGITAL world Chair Klaus HAUPT (REA), Moderator Nada MILISAVLJEVIC (DG HOME)

14:30-14:45 Blockchain: a new framework for swarm RObotic Systems (H2020-MSCA-IF-2016 BROS 751615) Eduardo CASTELLÓ FERRER, Université Libre De Bruxelles

14:45-15:00 AdvanCed Hardware/Software Components for Integrated/Embedded Vision SystEms (H2020-MSCA-ITN-2017 ACHIEVE 765866), Ricardo CARMONA-GALÁN, CSIC-Univ. Seville

15:00-15:15 High-Accuracy Indoor Tracking and Augmented Sensing using Swarms of UAVs (H2020-MSCA-IF-2017 AirSens 793581), Anna GUERRA, UNIBO and Stony Brook University

15:15-15:30 EnhaNcing seCurity And privacy in the Social wEb: a user centered approach for the protection of minors (H2020-MSCA-RISE-2015 ENCASE 691025), Michael SIRIVIANOS, Cyprus University of Technology

15:30-16:00 Panel discussion, Q&A

16:00-16:30 *Coffee break (COVE A2 5th floor auditorium)

Panel 2 Artificial Intelligence in DIGITAL world Chair Martin MUEHLECK (DG EAC), Moderator Cécile HUET (DG CNECT)

16:30-16:45 The use of computational techniques to Improve compliance to reminders within smart environments (H2020-MSCA-RISE-2016 REMIND 734355), Christopher NUGENT, Ulster University

16:45-17:00 Applications of Personal Robotics for Interaction and Learning (H2020-MSCA-ITN-2015 APRIL 674868), Pontus LOVIKEN, University Of Plymouth

17:00-17:15 Donating Robots a Theory of Mind (H2020-MSCA-IF-2014 DoRoThy 657227), Séverin LEMAIGNAN, Bristol Robotics Laboratory

17:15-17:30 Adaptive Decision support for Agents negotiation in electricity market and smart grid Power Transactions (H2020-MSCA-IF-2015 ADAPT 703689), Tiago PINTO, Polytechnic of Porto


18:00-18:30 Artificial Intelligence: EU funding in H2020 and beyond, Cécile HUET (DG CNECT)


44

18:30-20:30 *Poster session (drink and buffet - COVE A2 5th floor auditorium)

DAY 2 – Wednesday 11 December (COVE A2 4th floor auditorium)

Panel 3 Towards Trustworthy Artificial Intelligence Chair Julie LEPRETRE (DG EAC), Moderator Irina REYES (DG R&I)

09:00-09:15 Artificial Intelligence and Ethics, Charlotte STIX (DG CNECT) 09:15-09:30 Artificial Intelligence in Education: Ethical Aspects, Marcelino CABRERA GIRALDEZ (JRC) 09:30-09:45 Artificial Intelligence and Security, Nada MILISAVLJEVIC (DG HOME) 9:45-10:15 Panel discussion, Q&A

10:15-10:45 *Coffee Break (COVE A2 5th floor auditorium)

Panel 4 Artificial Intelligence in HEALTH Chair Jean-Bernard VEYRET (REA), Moderator Karim BERKOUK (DG R&I)

10:45-11:00 Rapid Biomechanics Simulation for Personalized Clinical Design (H2020-MSCA-ITN-2017 Rainbow 764644), Sune Darkner, Kobenhavns Universitet

11:00-11:15 A pEdiatRic dosimetRy personalized platfORm based on computational anthropomorphic phantoms (H2020-MSCA-RISE-2015 ERROR 691203), Panagiotis PAPADIMITROULAS, Bioemission Technology Solutions IKE

11:15-11:30 Automated Measurement of Engagement Level of Children with Autism Spectrum Conditions during Human-robot Interaction (H2020-MSCA-IF-2015 EngageME 2 701236), Ognjen Oggi RUDOVIC, Universitaet Augsburg

11:30-11:45 A Biomimetic Learning Control Scheme for control of Modular Robots (H2020-MSCA-IF-2015 BIOMODULAR 705100), Silvia TOLU, Danmarks Tekniske Universitet


12:15-13:30 *Lunch (COVE A2 5th floor auditorium)

Panel 5 Artificial Intelligence in HEALTH Chair Fredrik OLSSON HECTOR (REA), Moderator Yiannos TOLIAS (DG GROW)

13:30-13:45 Microwave Diagnosis of Breast Cancer with Open Ended Contact Probes (H2020-MSCA-IF-2016 MIDxPRO 750346), Tuba YILMAZ ABDOLSAHEB, Istanbul Teknik Universitesi

13:45-14:00 Training Network on Automatic Processing of PAthological Speech (H2020-MSCA-ITN-2017 TAPAS 766287), Mathew MAGIMAI DOSS, Fondation De L'institut De Recherche IDIAP

14:00-14:15 Outcomes of Patients’ Evidence With Novel, Do-It-Yourself Artificial Pancreas Technology (H2020-MSCA-RISE-2018 OPEN 823902), Shane O'DONNELL, University College Dublin, National University Of Ireland

14:15-14:30 Deep Understanding of Patient Experience of Healthcare from Social Media (H2020-MSCA-IF-2017 Deep Patient 794196) Lin GUI, Aston University


15:00-15:30 *Coffee break (COVE A2 5th floor auditorium)

Panel 6 Artificial Intelligence in ENVIRONMENT Chair Klaus-Günther BARTHEL (REA), Moderator Joachim D'EUGENIO (DG ENV)

15:30-15:45 Deep Learning UAV Networks for Autonomous Forest Firefighting (H2020-MSCA-IF-2016 DUF 752669), Nazim Kemal URE, Istanbul Teknik Universitesi

15:45-16:00 Probabilistic neuromorphic architecture for real-time Sensor fusion applied to Smart, water quality monitoring systems (H2020-MSCA-IF-2016 ProbSenS 747848) Josep Maria MARGARIT-TAULÉ, Universitat Zurich

16:00-16:15 Internet of Thing for Smart Water Innovative Networks (H2020-MSCA-ITN-2017 IoT4Win 765921), Wenyan WU, Birmingham City University

16:15-16:30 Low Energy Dew Point Cooling for Computing Data Centres


45

(H2020-MSCA-RISE-2016 DEW-COOL-4-CDC 734340), Yousef AKHLAGHI, University Of Hull 16:30-17:00 Panel discussion, Q&A

17:00-17:30 Closing remarks and wrap-up of event Fredrik OLSSON HECTOR (REA)

*Poster presentations (COVE A2 5th floor auditorium)

1. ACCWA Accounting for Climate Change in Water and Agriculture management (H2020-MSCA-RISE-2018-823965), Maria José ESCORIHUELA, isardSAT SL

2. DREAM-GO Enabling Demand Response for short and real-time Efficient And Market Based smart Grid Operation - An intelligent and real-time simulation approach (H2020-MSCA-RISE-2014-641794), Tiago PINTO, Instituto Politecnico Do Porto

3. ECOBOTICS.SEA Bio-inspired Technologies for a Sustainable Marine Ecosystem (H2020-MSCA-RISE-2018- 824043), Jorge DIAS, Instituto De Sistemas E Robotica-Associacao

4. ENDORSE Safe, Efficient and Integrated Indoor Robotic Fleet for Logistic Applications in Healthcare and Commercial Spaces (H2020-MSCA-RISE-2018-823887), Nacim RAMDANI, Universite D'Orléans

5. GREENDC Sustainable energy demand side management for GREEN Data Centers (H2020-MSCA-RISE-2016-734273), Habin LEE, Brunel University London and Tuba GÖZEL, Gebze Technical University

6. HYBSPN Hybrid Learning Systems utilizing Sum-Product Networks (H2020-MSCA-IF-2017-797223), Robert PEHARZ, The Chancellor Masters And Scholars Of The University Of Cambridge (UCAM)

7. IDENTITY Computer Vision Enabled Multimedia Forensics and People Identification (H2020-MSCA-RISE-2015-690907), Chang-Tsun LI, University Of Warwick

8. LAMBDA Learning and Analysing Massive/Big complex Data (H2020-MSCA-RISE-2016-734242), Nicholas ZYGOURAS, Ethniko Kai Kapodistriako Panepistimio Athinon

9. MetaPlat, Development of an Easy-to-use Metagenomics Platform for Agricultural Science (H2020-MSCA-RISE-2015-690998), Huiru ZHENG, University Of Ulster

10. NEPSpiNN Neuromorphic EMG Processing with Spiking Neural Networks (H2020-MSCA-IF-2016-753470), Elisa DONATI, Universitat Zurich

11. PRO GAIT Physiological and Rehabilitation Outcomes: Gains from Automated Interventions in stroke Therapy (H2020-MSCA-RISE-2017-778043), Olive LENNON, University College Dublin, National University Of Ireland

12. RUC-APS Enhancing and implementing Knowledge based ICT solutions within high Risk and Uncertain Conditions for Agriculture Production Systems (H2020-MSCA-RISE-2015-691249), Jorge HERNANDEZ, University of Liverpool

13. SENSIBLE SENSors and Intelligence in BuLt Environment (H2020-MSCA-RISE-2016-734331), Lina STANKOVIC, Vladimir STANKOVIC, University of Strathclyde

14. SKELETON-ID Soft Computing and Computer Vision for Comparative Radiography in Forensic (H2020-MSCA-IF-2016-746592), Pablo MESEJO, Universidad De Granada

15. SME 4.0 Industry 4.0 for SMEs - Smart Manufacturing and Logistics for SMEs in an X-to-order and Mass Customization Environment (H2020-MSCA-RISE-2016-734713), Erwin RAUCH, Libera Universita Di Bolzano

16. TIC-AUV Towards Intelligent Cognitive AUVs (H2020-MSCA-IF-2015-709136), Francesco MAURELLI, Jacobs University Bremen GGMBH

17. Water4Cities Holistic Surface Water and Groundwater Management for Sustainable Cities (H2020-MSCA-RISE-2016-734409), Dimitris KOFINAS, University of Thessaly


46

Annex 2: Participating MSCA projects9

ACCWA* Accounting for Climate Change in Water and Agriculture management. H2020-MSCA-RISE-2018-823965.

ACHIEVE AdvanCed Hardware/Software Components for Integrated/Embedded Vision SystEms. H2020-MSCA-ITN-2017-765866. ADAPT Adaptive Decision support for Agents negotiation in electricity market and

smart grid Power Transactions. H2020-MSCA-IF-2015-703689. AirSens High-Accuracy Indoor Tracking and Augmented Sensing using Swarms

of UAVs. H2020-MSCA-IF-2017-793581. APRIL Applications of Personal Robotics for Interaction and Learning. H2020-MSCA-ITN-2015-674868.

BIOMODULAR A Biomimetic Learning Control Scheme for control of Modular Robots. H2020-MSCA-IF-2015-705100.

BROS Blockchain: a new framework for swarm RObotic Systems H2020-MSCA-IF-2016-751615. Deep Patient Deep Understanding of Patient Experience of Healthcare from

Social Media. H2020-MSCA-IF-2017-794196. DEW-COOL-4-CDC Low Energy Dew Point Cooling for Computing Data Centres.

H2020-MSCA-RISE-2016-734340. DoRoThy Donating Robots a Theory of Mind. H2020-MSCA-IF-2014-657227. DREAM-GO* Enabling Demand Response for short and real-time Efficient And

Market Based smart Grid Operation - An intelligent and real-time simulation approach. H2020-MSCA-RISE-2014-641794.

DUF Deep Learning UAV Networks for Autonomous Forest Firefighting. H2020-MSCA-IF-2016-752669. ECOBOTICS.SEA* Bio-inspired Technologies for a Sustainable Marine

Ecosystem. H2020-MSCA-RISE-2018- 824043. ENCASE EnhaNcing seCurity And privacy in the Social wEb: a user centered

approach for the protection of minors. H2020-MSCA-RISE-2015-691025. ENDORSE* Safe, Efficient and Integrated Indoor Robotic Fleet for Logistic Applications in Healthcare and Commercial Spaces. H2020-MSCA-RISE-2018-

823887. EngageME 2 Automated Measurement of Engagement Level of Children with

Autism Spectrum Conditions during Human-robot Interaction. H2020-MSCA-IF-2015-701236.

ERROR A pEdiatRic dosimetRy personalized platfORm based on computational anthropomorphic phantoms. H2020-MSCA-RISE-2015-691203. GREENDC* Sustainable energy demand side management for GREEN Data

Centers. H2020-MSCA-RISE-2016-734273.

9 *Posters


47

HYBSPN* Hybrid Learning Systems utilizing Sum-Product Networks. H2020-

MSCA-IF-2017-797223. IDENTITY* Computer Vision Enabled Multimedia Forensics and People

Identification. H2020-MSCA-RISE-2015-690907. IoT4Win Internet of Thing for Smart Water Innovative Networks. H2020-MSCA-ITN-2017-765921.

LAMBDA* Learning and Analysing Massive/Big complex Data. H2020-MSCA-RISE-2016-734242.

MetaPlat* Development of an Easy-to-use Metagenomics Platform for Agricultural Science. H2020-MSCA-RISE-2015-690998. MIDxPRO Training Network on Automatic Processing of PAthological Speech.

H2020-MSCA-ITN-2017 TAPAS-766287. OPEN Outcomes of Patients’ Evidence With Novel, Do-It-Yourself Artificial

Pancreas Technology. H2020-MSCA-RISE-2018-823902. PRO GAIT* Physiological and Rehabilitation Outcomes: Gains from Automated

Interventions in stroke Therapy. H2020-MSCA-RISE-2017-778043. ProbSenS Probabilistic neuromorphic architecture for real-time Sensor fusion applied to Smart, water quality monitoring systems. H2020-MSCA-IF-2016-

747848. Rainbow Rapid Biomechanics Simulation for Personalized Clinical Design. H2020-

MSCA-ITN-2017-764644. REMIND The use of computational techniques to Improve compliance to reminders within smart environments. H2020-MSCA-RISE-2016-734355.

RUC-APS* Enhancing and implementing Knowledge based ICT solutions within high Risk and Uncertain Conditions for Agriculture Production Systems. H2020-

MSCA-RISE-2015-691249. SENSIBLE* SENSors and Intelligence in BuLt Environment. H2020-MSCA-RISE-2016-734331.

SKELETON-ID* Soft Computing and Computer Vision for Comparative Radiography in Forensic. H2020-MSCA-IF-2016-746592.

SME 4.0* Industry 4.0 for SMEs - Smart Manufacturing and Logistics for SMEs in an X-to-order and Mass Customization Environment. H2020-MSCA-RISE-2016-734713.

TIC-AUV *Towards Intelligent Cognitive AUVs. H2020-MSCA-IF-2015-709136. Water4Cities* Holistic Surface Water and Groundwater Management for

Sustainable Cities. H2020-MSCA-RISE-2016-734409.


48

Annex 3: Meeting Contributors

Alain VERSCHOREN, Keynote speaker, University of Antwerp. Alessandra LUCHETTI, Head of Department A Excellent Science, REA.

Anna GUERRA, UNIBO and Stony Brook University. H2020-MSCA-IF-2017-793581 AirSens. Cécile HUET, Deputy Head of the Unit Robotics and Artificial Intelligence, DG

CNECT. Chang-Tsun LI, University Of Warwick. H2020-MSCA-RISE-2015-690907

IDENTITY*. Charlotte STIX, Coordinator of the High-Level Expert Group on AI, DG CNECT. Christopher NUGENT, Ulster University. H2020-MSCA-RISE-2016-734355

REMIND. Claire MOREL, Head of Unit C2 Marie Skłodowska-Curie Actions, DG EAC.

Dimitris KOFINAS, University of Thessaly. H2020-MSCA-RISE-2016-734409 Water4Cities*. Eduardo CASTELLÓ FERRER, Université Libre De Bruxelles. H2020-MSCA-IF-

2016-751615 BROS. Elisa DONATI, Universitat Zurich. H2020-MSCA-IF-2016-753470 NEPSpiNN*.

Erwin RAUCH, Libera Universita Di Bolzano. H2020-MSCA-RISE-2016-734713 SME 4.0*. Francesco MAURELLI, Jacobs University Bremen GGMBH. H2020-MSCA-IF-2015-

709136 TIC-AUV*. Habin LEE, Brunel University London. H2020-MSCA-RISE-2016-73427

GREENDC*. Huiru ZHENG, University Of Ulster. H2020-MSCA-RISE-2015-690998 MetaPlat*. Irina REYES, Policy Assistant to the Director for Prosperity, DG R&I.

Joachim D'EUGENIO, Deputy Head of Division Compliance and Better Regulation, DG ENV.

Jorge DIAS, Instituto De Sistemas E Robotica-Associacao. H2020-MSCA-RISE-2018- 824043 ECOBOTICS.SEA*. Jorge HERNANDEZ, University of Liverpool. H2020-MSCA-RISE-2015-691249

RUC-APS*. Josep Maria MARGARIT-TAULÉ, Universitat Zurich. H2020-MSCA-IF-2016-747848

ProbSenS. Karim BERKOUK, Deputy Head of Unit Combatting diseases, DG R&I.

Lin GUI, Aston University. H2020-MSCA-IF-2017-794196 Deep Patient. Lina STANKOVIC, Vladimir STANKOVIC, University of Strathclyde. H2020-MSCA-RISE-2016-734331 SENSIBLE*.

Marcelino CABRERA GIRALDEZ, Senior researcher, European Commission, JRC. Maria José ESCORIHUELA, isardSAT SL. H2020-MSCA-RISE-2018-823965

ACCWA*. Mathew MAGIMAI DOSS, Fondation De L'institut De Recherche IDIAP. H2020-MSCA-ITN-2017-766287 TAPAS.

Michael SIRIVIANOS, Cyprus University of Technology. H2020-MSCA-RISE-2015 -691025 ENCASE.


49

Nacim RAMDANI, Universite D'Orléans. H2020-MSCA-RISE-2018-823887

ENDORSE*. Nada MILISAVLJEVIC, Policy Officer, Innovation and Industry for Security, DG

HOME. Nazim Kemal URE, Istanbul Teknik Universitesi. H2020-MSCA-IF-2016-752669 DUF.

Nicholas ZYGOURAS, Ethniko Kai Kapodistriako Panepistimio Athinon. H2020-MSCA-RISE-2016-734242 LAMBDA.

Ognjen Oggi RUDOVIC, Universitaet Augsburg. H2020-MSCA-IF-2015-701236 EngageME 2. Olive LENNON, University College Dublin, National University Of Ireland. H2020-

MSCA-RISE-2017-778043 PRO GAIT*. Pablo MESEJO, Universidad De Granada. H2020-MSCA-IF-2016-746592

SKELETON-ID*. Panagiotis PAPADIMITROULAS, Bioemission Technology Solutions IKE. H2020-

MSCA-RISE-2015-691203 ERROR. Pontus LOVIKEN, University Of Plymouth. H2020-MSCA-ITN-2015-674868 APRIL. Ricardo CARMONA-GALÁN, CSIC-Univ. Seville. H2020-MSCA-ITN-2017-765866

ACHIEVE. Robert PEHARZ, The Chancellor Masters And Scholars Of The University Of

Cambridge. H2020-MSCA-IF-2017-797223 HYBSPN*. Séverin LEMAIGNAN, Bristol Robotics Laboratory. H2020-MSCA-IF-2014-657227 DoRoThy.

Shane O'DONNELL, University College Dublin, National University Of Ireland. H2020-MSCA-RISE-2018-823902 OPEN.

Silvia TOLU, Danmarks Tekniske Universitet. H2020-MSCA-IF-2015-705100 BIOMODULAR. Sune Darkner, Kobenhavns Universitet. H2020-MSCA-ITN-2017-764644 Rainbow.

Tiago PINTO, Instituto Politecnico Do Porto. H2020-MSCA-RISE-2014-641794 DREAM-GO*.

Tiago PINTO, Polytechnic of Porto. H2020-MSCA-IF-2015-703689 ADAPT. Tuba GÖZEL, Gebze Technical University. H2020-MSCA-RISE-2016-73427 GREENDC*.

Tuba YILMAZ ABDOLSAHEB, Istanbul Teknik Universitesi. H2020-MSCA-IF-2016-750346 MIDxPRO.

Wenyan WU, Birmingham City University. H2020-MSCA-ITN-2017-765921 IoT4Win. Yiannos TOLIAS, Lawyer in AI, liability and market surveillance for products, DG

GROW. Yousef AKHLAGHI, University Of Hull. H2020-MSCA-RISE-2016-734340 DEW-

COOL-4-CDC.


50

The Marie Skłodowska-Curie actions (MSCA)

projects are managed by the

Research Executive Agency (REA)

Follow REA on Twitter: @REA_research

https://ec.europa.eu/info/departments/research-executive-agency_en

https://twitter.com/REA_research

H2020 Marie Skłodowska-Curie Actions Artificial Intelligence ...

Documents