BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Special BMI Seminar // Rikki Rabinovich & Jack Bowler
DTSTART:20250827T160000
DTEND:20250827T180000
DTSTAMP:20260428T030708Z
UID:7e885e0755e814d6b02c6ca03e9ace116d09d799a89e2e1757ec44b4
CATEGORIES:Conferences - Seminars
DESCRIPTION:Rikki Rabinovich & Jack Bowler\, University of Utah\, Departm
 ent of Neurosurgery\, USA\nSpeaker 1: Rikki Rabinovich (University of Utah
 \, Department of Neurosurgery)\nTitle: Human brain network control of crea
 tivity\n\nThe ability to think creatively is fundamental to human cognitio
 n\, shaping how we interact with the world and navigate complex problems. 
 In this talk\, I discuss our recent findings that reveal how the human bra
 in generates creative ideas. Using multiscale intracranial electrophysiolo
 gy—from single-unit recordings to local field potentials across distribu
 ted brain networks—we demonstrated that unique brain states underlie cre
 ative vs. mathematical cognitive states. Further\, we uncovered nonlinear\
 , high-dimensional representations of moment-by-moment fluctuations in cre
 ative performance. These representations define a creativity axis shared b
 etween network-level and single-neuron computations. Our findings reveal w
 idespread neural representations of cognitive state and suggest distinct r
 oles of specific brain networks in controlling creativity\, with the defau
 lt mode network gating access to creative states and the dorsal attention 
 network regulating the quality of creative output.\n\nSpeaker 2: Jack Bowl
 er (University of Utah\, Department of Neurobiology)\nTitle: Learning to g
 eneralize: Artificial networks reveal how the structure of prior experienc
 e shapes abstract thinking in real brains\n\nAnimals can solve new\, compl
 ex tasks by reusing and adapting what they’ve learned before. This kind 
 of flexibility depends not just on having prior experience\, but on how th
 at experience was structured in the first place. To investigate how early 
 training shapes learning\, we first trained recurrent neural networks (RNN
 s) on an odor-timing task previously used to study complex timing behavior
  in mice. We then tested the RNN predictions regarding the effect of exper
 ience on generalization by comparing simulated RNN results to behavioral a
 nd electrophysiological recordings in mice trained on the same task throug
 h a progression of staged training sequences. Both RNNs and mice trained w
 ithout well-structured early experience developed rigid strategies and mad
 e repeated errors. In contrast\, properly structured early training improv
 ed generalization in RNNs and mice. Using dynamical systems approaches\, w
 e revealed a mechanism for this effect: networks trained with appropriatel
 y structured curricula developed distinct dynamical motifs that support th
 e correct abstractions when complexity was increased. Networks that lacked
  early training or received improperly structured curricula developed sing
 le fixed-point solutions that failed to generalize about the task structur
 e. Together\, these findings demonstrate how the structure of prior experi
 ence governs flexible and generalizable knowledge emergence in both biolog
 ical systems and computational models.\n 
LOCATION:SV 2510 https://plan.epfl.ch/?room==SV%202510
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR