SMIRP Barnett 2002

SMIRP

Jean-Claude [email protected]

Drexel University

Barnett International Conference on Laboratory Notebooks

08/14/2002

Effective use of a self-evolving database for information capture and retrieval in an R&D

environment

http://smirp.drexel.edu

LIMS CENS

Single Instrument Automation

Laboratory Information Management Systems

Collaborative Electronic Notebook Systems

Human /Autonomous Agent Hybrid Systems

Human ManagedFully AutonomousScientific Research Systems

TODAY

SMIRP bridge

The Evolution of Automation in Scientific Research

HumanAgent

AutonomousAgent

SMIRP

Automation SWAT team

(Bot)

Browser

Excel

The SMIRP model for a hybrid Human/Autonomous Agent System

Anthropomimetic Design

Approaches to Collaborative Electronic Notebooks

rigid

SMIRPcompromise:

Rigid information representationFlexible linking of modules

flexible

•Structured•Generally

domainspecific

•Adaptable•Unstructured

http://smirp.drexel.edu

Add informationto database

Retrieveinformation

Modify database structure

Functional Requirements of a collaborative electronic notebook

SMIRPRequest

structuralmodification

http://smirp.drexel.edu

Two approaches to the development of databases

Communicateanticipated

need

Designdatabase structure

Let database structureevolve

through useSMIRP

http://smirp.drexel.edu

Fundamental Information Representation in SMIRP

Module 1 Module 2

Parameter 1

Parameter 2

Parameter 4

Parameter 5

instance

Record 1

instance

Record 2

http://smirp.drexel.edu

(People)

(Name)

(Employee of)

(Company)

(Name)

Parameter 3(email)

(Address)

Bill Gates Microsoft

Case-study: Evolution of SMIRP structure in a chemistry laboratory

Location Drexel University

Department of Chemistry

Users faculty, undergraduate students, graduatestudents, librarians and other university personnel

Period Feb 1999 – April 2001, with a detailed focus on

last 7 months (Sept 2000-April 2001)

Total accounts (last 7 months) 78

Active Accounts (added records) 50

Administrators (changed database structure)

9

http://smirp.drexel.edu

HumanResource

Management 13%

Maintenance1%

Knowledge Processing

72%

Most Active Module Categories (9/00 – 4/01)

Labwork14%

118 modules 1/3 account for 98% of activity

http://smirp.drexel.edu

Most Active Knowledge Processing Modules

Journal 9%

Knowledge Filter 3%

ReformatReference requests

20%FindReference

66%

PublisherDocument ProductionReference ProcessingParameter CorrelationData source filesExperimental Conclusion GenerationKnowledge consolidation

http://smirp.drexel.edu

Most Active Laboratory Modules

Preparation of Silver rods for SCBETEM Micrographs Of Pd on CSCBE on membranesHydrogenation of Crotonaldehyde using Pd CatalystsReduction of Methylene blue by Pd Metal Particles in a Field

Electrodeposition of Pd on Graphite

29%

Protocol Prototyping25%

Pd onto Carbon Nanofibers

17%

Electroless plating on Membranes

9%

Synthesis of Pd catalysts by Bipolar electrochemistry

5%

TEM Micrographs Of Pd on C

3%

Pd particle size analysis using TEM

3%

http://smirp.drexel.edu

Recruitment events 2%

ProjectManager

5%Errors5%

Productivity Tracking

14%

People 28%

Workstudy hours reporting

46%

Most Active Human Resource Management Modules

http://smirp.drexel.edu

Most Active Maintenance Modules

SMIRPProblems

22%

Orders 19%

Invoice (TEM/SEM and other instrument charges)

19%

Laboratorymaterials

16%

Vendor15%

Orderforms9%

http://smirp.drexel.edu

Activity Analysis by Category over Time

20

00

-10

-3

20

00

-10

-17

20

00

-10

-30

20

00

-11

-12

20

00

-11

-25

20

00

-12

-8

20

00

-12

-21

20

01

-1-3

20

01

-1-1

6

20

01

-1-3

0

20

01

-2-1

2

20

01

-2-2

5

20

01

-3-1

0

20

01

-3-2

3

20

01

-4-5

20

01

-4-1

8

Maintenance

Human Resource Management

Laboratory Work

Knowledge Processing0

1000

2000

3000

4000

5000

6000

7000

8000

http://smirp.drexel.edu

For agents to make a decision to:

ACT NOT ACT

Generally for quality controlExpected information: Retrieve details and execute from a menu of predefined tasks

Unexpected information: Redesign tasks

This could be absence of information:“No News is Good News”

WHY retrieve information?

Active

Passive

Negative (implied)

Pre-emptive

I want to know something NOW

Keep me updated regularly with new information

No news is good news

Tell me things I SHOULDwant to know but have not asked for

Burden on Agent

Highest

Lowest

Are your closest familymembers alive?

A competitor has initiated research in my market space

New experiments in a particular project

Obtaining a phone number

Description ExampleMode

HOW Agents Retrieve Information

Active

E-mail

Browser

Excel

Interface

Information Filter

TimeKeyword User ContextSimilarity

Passive

Operation

SMIRP Information Retrieval Matrix

Keyword Search Results: example “nanotube”Active Information Retrieval : keywords

From Keyword to Article

From Keyword to Knowledge Filter

From Keyword to Orders

From Keyword to Protocol Prototyping

Active Information Retrieval : Time Filtered

Active Information Retrieval: User and Operation Filtered Search

Autonomous AgentMonitoring

Active Information Retrieval: Similarity Based

Active Information Retrieval: Context Based

Passive Information Retrieval: Email Alerts

Space Level Module Level

All Activity

New EntriesWhen link to article has been foundMonitor progress of software development

Keep track of which software version users have downloadedMonitor which experiments are being investigated

Keep track of special users:Job applicantsFormer usersCollaborators

Updates on report or article being written

(general) (specific)

New activity related to keywords

Quality control of autonomous agent activity

Quality Control of workflows

Module-Level Alerts: Creation of an alert for new urls to articles

Module Level Alerts: Creation of an alert for new urls to articles

Space Level Alerts: example of keyword filtering

Seamless Integration of Human and Autonomous Agents in Workflows

Real-Time Workflow Designs

Automated

Human(default)

State A State B

Workflow for Extraction of Article information and url

Queries Web and extracts information

AutonomousAgent

Successful Processing

Citation to be Processed

Portal Not Found

Citation Invalid

Information Missing

HumanAgent Handles exceptions

Human/Autonomous Agent Coordination in Workflow

Pre-emptive information retrieval

Report experimental results

Read experimental results

Generate Search Strings

Read search strings

Report on search results

Alerted to new documents of potential interest

Parsebot Googlebot

Finding documents that should be of interest to current work

Pre-emptive information retrieval

Finding documents that should be of interest to current work

Pre-emptive information retrieval

Pre-emptive information retrieval

Finding documents that should be of interest to current work

Leveraging and Extending Bot Implementation

Citation bot in other laboratory research and teaching spaces

In online class SMIRPspace: Plagiobot System

Automatic Content Summarization Tools

Analysis/verification of experimental data analysis

Conversion from Passive to Negative Information Mode: Bot Monitoring of other Bots

Monitoring of competitor/collaborator activity (patents/papers)

Automatic Keyword Generation from most frequently used or read words

Conclusions

This is still a “Human World”

SMIRP can serve as a framework to allow Human and Autonomous Agents to operate freely within a Laboratory Research Collaborative Space

Automation within workflows can be accelerated by creating Autonomous Agents that are more Human-like in how they retrieve and store information

Acknowledgements

Benjamin SamuelSundar Babu

Raj HooliKetan Patel

Mohammad Haghkar

NSF CAREER CHE-9875855CIA

http://smirp.drexel.edu