AI, ORD & the Research Data Connectome

Prof. Dr. Philippe Cudré-Mauroux

eXascale Infolab, University of Fribourg

​

swissuniversities

March 30, 2022

Express Bio

• I build critical infrastructures for data
• Past: EPFL (best thesis award), Eurecom, INRIA, Berkeley, MIT
• Industry: HP, IBM Watson, Microsoft (Asia, Silicon Valley)
• Present: UNIFR, SNF, CHIST-ERA

​

• Research
• Knowledge Graphs (2001-2022) [EPFL]
• Big Data infrastructures (2008-2022) [MIT]
• AI (2013-2022) [UNIFR]
• Hardware/software co-design (2019-2022) [UNIFR]

New lab @ UNIFR

https://exascale.info/

Data is the new Oil

Data is the new oil

​

Data + Algorithms 🡺 Actionable Insight 🡺

Big Data /

Data Science

Machine Learning /

“Dumb” A.I.

Model

(Regression /

Classification)

New Products /

New Knowledge /

Optimized Services

Let’s talk about AI

AI ⊃ ML ⊃ DL

Artificial Intelligence

Intelligence is the goal�(nothing intelligent yet)

Machine Learning

Builds a statistical�model from data (mostly)

Deep Learning

Uses deep neural�networks for the model

Model Training: cat or not?

cat

not!

cat

=> cat!

cat

Deep Learning from data solves many tasks today

Deep Learning can beat humans on vertical tasks (1/2)

Deep Learning can beat humans on vertical tasks (2/2)

AI Biggest Threat?

• Artificial General Intelligence (AGI, strong AI), surely?

We are nowhere near AGI

• Zero consensus from experts today
• None of the techniques we developed so far exhibits “intelligence”
• My guess? AGI will come… eventually
• Yann LeCun: a few decades
• Geoffrey Hinton: nowhere close to reality
• Yoshua Bengio: nobody knows

​

That Being Said, Opportunities Are….

• … endless
• Future is bright for AI/ML
• Make sense of Big Data
• Automate/optimize any domain where repeated evaluation is possible and cheap

�

​

EU [+CH] needs to invest into AI… NOW

​

​

Autodetermination?

​

ORD & AI

• ORD infrastructures are rapidly getting outdated, esp. in CH
• Lagging Big Data / cloud infrastructures
• Nothing FAIR, outdated DMPs
• CH is increasingly isolated

​

• Dire need for new solutions
• Researchers depend on it!
• ORD & AI are made for each other!

​

​

ORD in 2022: Are we there yet?

• You are interested in electricity time series
• How do you find such datasets today?

Solution: the Research Data Connectome

“The Connectome connects and organizes (open) scientific (meta)data sustainably across disciplines to make it widely accessible, interoperable and valuable.”

​

​

Connectome - Overview

rescs.org

AskReco

Discovery

AI

KG

Project Timeline

19

Some Services

Linked Data Pipeline

• Data Linkage Service
• Data Transformation Service
• Data Enrichment Service

​

Connectome API

• Linked ORD Access

​

Connectome AI

• Insights as a Service (via API)

​

End-User Services

• Discovery Solutions

20

Recent progress

• Application ”Roadmap Research Infrastructures”.

​

• Large Scale data imports (3.8M).

​

• Move towards implementation governance & business model.

​

• New Data Transformations & Data Enrichments (Services).

​

• Cooperative Development of new use-cases incl. Insights.

​

• New collaborations with BFH, EPFL, FHGR, SLSP started.

21

AI, ORD and the Connectome

• New applications to improve data quality.
• Named Entity Extraction
• Disambiguation
• Text Extractions from images
• Etc..

​

• Generation of insights on ORD through
• abstract summarizations
• abstract simplifications
• Co-authorship network analysis
• Self-learning user interfaces
• …

​

22

Conclusions

• We are FAR from being FAIR…
• Urgent need to build critical infrastructures for data-driven research�
• AI can help organize large-scale ORD
• The Connectome is a critical step in the right direction

​

Thanks a lot for your attention!