BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//talks.cam.ac.uk//v3//EN
BEGIN:VTIMEZONE
TZID:Europe/London
BEGIN:DAYLIGHT
TZOFFSETFROM:+0000
TZOFFSETTO:+0100
TZNAME:BST
DTSTART:19700329T010000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0100
TZOFFSETTO:+0000
TZNAME:GMT
DTSTART:19701025T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
CATEGORIES:Computational Neuroscience
SUMMARY:Computational Neuroscience Journal Club - Jean-Pas
cal Pfister and Xizi Li
DTSTART;TZID=Europe/London:20210518T150000
DTEND;TZID=Europe/London:20210518T163000
UID:TALK160606AThttp://talks.cam.ac.uk
URL:http://talks.cam.ac.uk/talk/index/160606
DESCRIPTION:Please join us for our fortnightly journal club on
line via zoom where two presenters will jointly pr
esent a topic together. The next topic is ‘Nonline
ar filtering in neuroscience’ presented by Jean-Pa
scal Pfister and Xizi Li.\n\nZoom information: htt
ps://us02web.zoom.us/j/84958321096?pwd=dFpsYnpJYWV
NeHlJbEFKbW1OTzFiQT09\n\nContinuously extracting r
elevant information from a stream of inputs is a k
ey machine learning problem with a wide range of a
pplications in neuroscience. Formally\, this probl
em - also known as nonlinear (Bayesian) filtering
- aims at estimating the posterior distribution (o
r filtering distribution) at time t of some latent
variable given all the inputs up to time t.\n\nTh
is nonlinear filtering theory can be applied in ne
uroscience from two different perspectives. Firstl
y\, nonlinear filtering can be seen as a data anal
ysis method where the task is to extract relevant
information from continuous neural recording (such
as continuously estimating the position of a rat
based on place cells activity or continuously esti
mating the intention of a patient in order to cont
rol a neuroprothesis). Secondly\, nonlinear filter
ing can be seen as a computational principle that
can be applied at different levels such as the beh
avioural level (e.g. continuously tracking the pos
ition of a prey)\, neuronal level (estimating the
causes of the inputs to a neural network) or even
single synapse level (e.g estimating the presynapt
ic membrane potential).\n\nIn the first part of th
is journal club\, we will review the theory of non
linear filtering with the formal solution given by
the Kushner-Stratonovic equation. For a tutorial
see [1]. We will highlight the limitation of this
formal solution in terms of practical applicabili
ty and describe the possible approximate solutions
. In the second part of the journal club we will d
iscuss one specific application of nonlinear filte
ring in the context of learning [2\,3] and highlig
ht the specific predictions of a synaptic learning
rule derived from this nonlinear filtering approa
ch.\n\nRefs:\n\n[1] Kutschireiter\, A.\, Surace\,
S. C.\, & Pfister\, J.-P. (2020). The Hitchhiker’s
guide to nonlinear filtering. Journal of Mathemat
ical Psychology\, 94\, 102307. http://doi.org/10.1
016/j.jmp.2019.102307\n\n[2] Aitchison\, L.\, Jegm
inat\, J.\, Menendez\, J. A.\, Pfister\, J.-P.\, P
ouget\, A.\, & Latham\, P. E. (2021). Synaptic pla
sticity as Bayesian inference. Nature Neuroscience
\, 24\, 565–571. \nhttp://doi.org/10.1038/s41593-0
21-00809-5\n\n[3] Jegminat\, J.\, & Pfister\, J.-P
. (2020). Learning as filtering: \nimplications fo
r spike-based plasticity. arXiv:2008.03198\n
LOCATION:Online on Zoom
CONTACT:Jake Stroud
END:VEVENT
END:VCALENDAR