BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IC Colloquium: Advancing System Models of Brain-Like Intelligence 
 via Integrative Benchmarking
DTSTART:20220428T100000
DTEND:20220428T110000
DTSTAMP:20260511T034918Z
UID:fb390a181d9af60afc80109205d69d270c11510392226a9b865e3f59
CATEGORIES:Conferences - Seminars
DESCRIPTION:By: Martin Schrimpf - MIT\nIC/SV Faculty candidate\n\nAbstract
 \nEfforts in machine learning attempt to engineer artificially intelligent
  models\, while research in the brain and cognitive sciences attempts to u
 ncover the neural circuitry underlying human intelligence in domains such 
 as vision and language. I will argue that the time is ripe (again) for ric
 h synergies between both fields: build system models that capture neural m
 echanisms and supported behaviors within entire domains of intelligence. T
 o make progress on system models\, I propose integrative benchmarking – 
 integrating experimental results from many laboratories into suites of ben
 chmarks that guide and constrain those models at multiple stages and scale
 s. We show-case this approach by developing Brain-Score benchmark suites f
 or neural and behavioral experiments in the primate visual ventral stream 
 as well as the human language system. By systematically evaluating a wide 
 variety of model candidates\, we not only identify models beginning to mat
 ch a range of brain data (~50% explained variance)\, but also discover key
  relationships: Models’ brain scores are predicted by their object categ
 orization performance in vision (but only up to 70% ImageNet accuracy)\, a
 nd their next-word prediction performance in language. The better models p
 redict internal neural activity\, the better they match human behavioral o
 utputs\, with architecture substantially contributing to brain-like repres
 entations. Using the integrative benchmarks\, we develop improved state-of
 -the-art system models that more closely match shallow recurrent neuroanat
 omy\, predict primate temporal processing\, and are more robust to image c
 orruptions. Taken together\, these integrative benchmarks and system model
 s are first steps to modeling the complexities of brain processing in enti
 re domains of intelligence.\n\nBio\nMartin is a research scientist at the 
 MIT Quest for Intelligence\, where he bridges research in natural and arti
 ficial intelligence. He focuses on computational models of vision and lang
 uage that are neurally and behaviorally aligned with primates. Martin got 
 his PhD from MIT\, working with James DiCarlo on the neural mechanisms und
 erlying object recognition. Before\, he worked on natural language process
 ing with Richard Socher at Salesforce Research\, and on human pattern comp
 letion with Gabriel Kreiman at Harvard. Martin got his Bachelor and Master
  degrees from TUM\, LMU\, and UNA in Germany. He co-founded the non-profit
  startup Integreat\, and previously worked at Siemens and Oracle Research.
 \n\nMore information
LOCATION:BC 420 https://plan.epfl.ch/?room==BC%20420 https://epfl.zoom.us/
 j/68589422153?pwd=WDBKQ25DckJzUERmRDl1TzlSU3Q3Zz09
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR