The main types of machine learning and their practical application

Main machine learning systems and their business usage

About me

Illarion Khlestov

Researcher at Ring Ukraine, computer vision department

GitHub: https://github.com/ikhlestov

Blog: https://medium.com/@illarionkhlestov

Facebook: https://www.facebook.com/i.khlestov

- Machine learning is just a tool.

- The tool that may help you and your business.

- ML may not be easy, but at least it’s possible.

- It’s interesting.

- And in any case ML is very popular.

Main ideas

Agenda

- Industry overview

- Closer look at:

- Chatbots

- Healthcare

- Autonomous driving

What is machine learning?

What is machine learning?

Industry Overview. What is the reason of ML?

- ML market - 1.41 Billion in the end of 2017.

- Expected on 2022 - 8.81 Billion (report)

- Company engaged:

- Toyota, VAG group, Daimler AG

- Walmart, Target, Amazon

- AIG, PayPal, Zappos

- ...

- How?

- Personalize

- Automate

- Predict

- Improve

- ...

Chatbots

The easiest bot

A little bit better example

ML solution

Available tools and approaches

Words to vectors

Word-to-vec example

You may try it online:

http://projector.tensorflow.org/

DialogFlow

Business Values

- Reduced costs

- Customers happiness

- Response rate

- 24/7 availability

- Scalability

- Additional training

What’s next? VoiceBots?

- Customers intention understanding

- Complicated actions

- Speech recognition

- Voice generation

Healthcare

What does exist now?

- Digital medical records

- Disease identification/Diagnosis

- Drugs discovery/Manufacturing

- Epidemic outbreak prediction

What can be done?

- Wearable continuous monitoring devices

- Single database

- Personalized medicine

- Automatic treatment or recommendation

- Automated handling of medical records

- Treatment of disabled people

- People modifications

How is it possible?

- Objects classification

- Objects detection

- Prediction systems

- Speech and text recognition

Business values

- Increased life expectancy

- Reduction of insurance payments

- Improvements in the one of the most huge markets

Potential problems

- Data availability

- Personal data handling and

protecting

- False positive or false negative

results

- Certification, medical clearance

- Bureaucracy and conservatism

Autonomous Driving

Current state of the field

Grounding

- Safety

- Traffic improvements

- Costs reducing

- Cargo transportation

Blockers

- Legal issues

- Opaque decision system

- People

- Privacy

- Other...

Adversarial Attack

Adversarial Attack

Adversarial Attack

Moral issues: what should car do?http://moralmachine.mit.edu/

Job losses

- Approximate 3.5 million of truck drivers

- Abt .5 million of taxi drivers

- Support staff

What is mainly used

- Objects detection

- Segmentation

- Tracking

- Reinforcement learning

- Usual SGD

- SLAM

SLAM - Simultaneous localization and mapping

Existed resources

- Udacity Self Driving Cars nanodegree

- Open Source Self Driving Car Initiative

- MIT 6.S094: Deep Learning for Self-Driving Cars

- Autonomous Driving CookBook

- Nvidia end-to-end training paper

General Overview

Are you need it?

- What benefit?

- What are implementation costs?

Take a look at the possible blockers:

- Is such task implementable with the help of ML at all?

- Legal issues

- Datasets existence

First steps:

- Consult with domain expert

- Define clear requirements(minimum and maximum)

- Speed

- Accuracy

- What should be considered as "done"?

- Check available open sourced solutions

Later:

- Measure real profit

- Decide, should your solution be updated or not

Thank you!Questions?

GitHub: https://github.com/ikhlestov

Blog: https://medium.com/@illarionkhlestov

Facebook: https://www.facebook.com/i.khlestov

UDS Community: https://www.facebook.com/groups/udsclub/

Bonus: another fields with ML

- Recommendation systems.

- Market analysis. Market prediction and targeting.

- Security systems.

- Content adjusting.

- Agriculture usage. Diseases detection, harvest prediction…

- Generative models. Routes planning, development and arts.

- Physical world modelling.

- Virtual Reality.