BEGIN:VCALENDAR
VERSION:2.0
PRODID:talks.ox.ac.uk
BEGIN:VEVENT
SUMMARY:A taxonomy of recurrence - Dr. Marcus Ghosh (Imperial College Lond
 on)
DTSTART;VALUE=DATE-TIME:20260423T150000
DTEND;VALUE=DATE-TIME:20260423T170000
UID:https://talks.ox.ac.uk/talks/id/d519939d-204e-4f9c-a51c-50975b69b805/
DESCRIPTION:How does the structure of a neural network shape its function?
  In this talk I will introduce partially recurrent neural networks (pRNNs)
 : a model in which a set of connection pathways can be combined combinator
 ially to generate a complete taxonomy of architectures between feedforward
  and fully recurrent. I will present two functional explorations across th
 ese structures. First\, using closed-form solutions\, I will demonstrate t
 hat linear pRNNs exhibit surprisingly diverse temporal dynamics\, includin
 g transient amplifications and oscillations\, which are approximately inva
 riant to network size. Second\, using nonlinear pRNNs trained with deep re
 inforcement learning\, I will show that distinct architectures differ in t
 heir learning speed\, peak performance\, and robustness to various perturb
 ations. I will conclude by mapping these functional differences to specifi
 c network traits\, illustrating how pRNNs can illuminate structure-functio
 n principles relevant to both neuroscience and machine learning.\nSpeakers
 :\nDr. Marcus Ghosh (Imperial College London)
LOCATION:Sherrington Library\, off Parks Road OX1 3PT
TZID:Europe/London
URL:https://talks.ox.ac.uk/talks/id/d519939d-204e-4f9c-a51c-50975b69b805/
BEGIN:VALARM
ACTION:display
DESCRIPTION:Talk:A taxonomy of recurrence - Dr. Marcus Ghosh (Imperial Col
 lege London)
TRIGGER:-PT1H
END:VALARM
END:VEVENT
END:VCALENDAR