BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20250628T134243EDT-09846UupHI@132.216.98.100
DTSTAMP:20250628T174243Z
DESCRIPTION:Neural dynamics shape task organization in multitask networks\n
 \nDavid Sussillo\, Stanford University)\n	Tuesday February 21\, 12-1pm\n	Zoo
 m Link: https://mcgill.zoom.us/j/86855481591\n\nAbstract: Flexible computa
 tion is a hallmark of intelligent behavior. Yet\, little is known about ho
 w neural networks contextually reconfigure for different computations. Hum
 ans are able to perform a new task without extensive training\, presumably
  through the composition of elementary processes that were previously lear
 ned. Cognitive scientists have long hypothesized the possibility of a comp
 ositional neural code\, where complex neural computations are made up of c
 onstituent components\; however\, the neural substrate underlying this str
 ucture remains elusive in biological and artificial neural networks. Here 
 we identified an algorithmic neural substrate for compositional computatio
 n through the study of multitasking artificial recurrent neural networks. 
 Dynamical systems analyses of networks revealed learned computational stra
 tegies that mirrored the modular subtask structure of the task-set used fo
 r training. Dynamical motifs such as attractors\, decision boundaries and 
 rotations were reused across different task computations. For example\, ta
 sks that required memory of a continuous circular variable repurposed the 
 same ring attractor. We show that dynamical motifs are implemented by clus
 ters of units and are reused across different contexts\, allowing for flex
 ibility and generalization of previously learned computation.\n
DTSTART:20230224T170000Z
DTEND:20230224T180000Z
SUMMARY:QLS Seminar Series - David Sussillo
URL:/qls/channels/event/qls-seminar-series-david-sussi
 llo-346055
END:VEVENT
END:VCALENDAR