BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computational Neuroscience Journal Club - Gido van de Jen
DTSTART:20201201T150000Z
DTEND:20201201T163000Z
UID:TALK154531@talks.cam.ac.uk
CONTACT:Jake Stroud
DESCRIPTION:Please join us for our fortnightly journal club online via zoo
 m where two presenters will jointly present a topic together.\n\nJoin Zoom
  Meeting\nhttps://us02web.zoom.us/j/89883818695?pwd=TTVVYitVT1VXMHZ5UXIwTF
 E4ZCtMQT09\n\nMeeting ID: 898 8381 8695\n\nPasscode: 668906\n\nThe next to
 pic is ‘computational models of synaptic complexity’ presented by Gido
  Van de Jen and Xizi Li.\n\nBiological synapses are highly complex with a 
 multitude of molecular signalling pathways. Yet\, in classical models of s
 ynaptic plasticity as well as in deep neural networks\, synaptic efficacy 
 is typically modelled as a single scalar value. Moreover\, theoretical con
 siderations alone diminish such representations\, as neural networks with 
 scalar synapses have strikingly limited memory capacity once you assume a 
 finite number of distinguishable levels of synaptic strengths. In this jou
 rnal club\, we shall first discuss theoretical studies using the “ideal 
 observer formalism” that show that the memory capacity of complex synaps
 es can be substantially higher than that of simple scalar synapses with th
 e realistic assumption of limited precision in synaptic strength ([1] and 
 related work). Then\, we explore whether synaptic complexity can be benefi
 cial in deep neural networks (DNNs). Specifically\, we introduce how the c
 omplex synapse model from [1] is applied “out-of-the-box” to reinforce
 ment learning with Deep Q-Networks [2]\, and how a slightly different vers
 ion of synaptic complexity substantially reduces catastrophic forgetting i
 n DNNs [3].\n\n[1] Benna\, M. K.\, & Fusi\, S. (2016). Computational princ
 iples of synaptic memory consolidation. Nature neuroscience 19(12): 1697-1
 706\, https://www.nature.com/articles/nn.4401\n\n[2] Kaplanis\, C.\, Shana
 han\, M.\, & Clopath\, C. (2018). Continual Reinforcement Learning with Co
 mplex Synapses. International Conference on Machine Learning\, https://arx
 iv.org/abs/1802.07239\n\n[3] Zenke\, F.\, Poole\, B.\, & Ganguli\, S. (201
 7). Continual learning through synaptic intelligence. International Confer
 ence on Machine Learning\, https://arxiv.org/abs/1703.04200
LOCATION:Online on Zoom
END:VEVENT
END:VCALENDAR