BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Reactivation in biological and artificial neural networks - Gido v
 an de Ven\, PhD\; Baylor College of Medicine
DTSTART:20180911T090000Z
DTEND:20180911T100000Z
UID:TALK109846@talks.cam.ac.uk
CONTACT:Yul Kang
DESCRIPTION:The brain’s ability to retain memories over a lifespan is th
 ought to rely on the reactivation of memory-representing cell assemblies. 
 To experimentally test this\, for my PhD I performed multi-unit recordings
  in the hippocampus of mice exploring novel environments. We found that se
 lectively disrupting reactivation by closed-loop optogenetic silencing of 
 sharp-wave/ripples impaired the later reinstatement of recently-formed pla
 ce cell assemblies\, thereby providing the first direct evidence that reac
 tivation stabilizes new cell assemblies.\nTo gain deeper insight into the 
 computational role of reactivation\, for my postdoc I turned to machine le
 arning. Current state-of-the-art artificial neural networks are my ideal 
 “model organism” as they can perform extremely well on a wide variety 
 of individual tasks\, but they struggle to retain old information when tra
 ined on new tasks. Could reactivation improve memory consolidation in arti
 ficial neural networks? To test this\, we equipped deep feed-forward netwo
 rks for classification with feedback connections trained to have generativ
 e capability. We found that interleaving “reactivation” generated by t
 hese feedback connections with new task data substantially reduced the cat
 astrophic forgetting of old tasks. Notably\, for classification of MNIST-d
 igits\, this approach outperforms and is more widely applicable then curre
 nt deep learning strategies for alleviating catastrophic forgetting. To fi
 nish\, I will discuss (1) why I believe this approach could scale to more 
 complicated inputs and (2) how I plan to use further insights from the bra
 in to achieve this.
LOCATION:Cambridge University Engineering Dept.\, Board Room (2nd floor\, 
 Baker building)
END:VEVENT
END:VCALENDAR