19 Mar
2024
19 Mar
'24
10:11 a.m.
Dear All,
The CEU Department of Cognitive Science and the Center for Cognitive Computation (CCC)
invites you to the following talk.
*Please note that this event will be held at CEU Budapest site (N13. building), from 5
PM.*
Speaker: Ádám Gosztolai (EPFL)
Title: Interpretable representations of neural dynamics using geometric deep learning
https://arxiv.org/abs/2304.03376
Abstract: It is increasingly recognised that computations in the brain and artificial
neural networks can be understood as outputs of a high-dimensional dynamical system
conformed by the activity of large neural populations. Yet revealing the structure of the
underpinning latent dynamical processes from data and interpreting their relevance in
computational tasks remains a fundamental challenge. A prominent line of research has
observed that task-relevant neural activity often takes place on low-dimensional smooth
subspaces of the state space called neural manifolds. However, there is a lack of
theoretical frameworks for the unsupervised representation of neural dynamics that are
interpretable based on behavioural variables, comparable across systems, and decodable to
behaviour with high accuracy.
To address these challenges, we introduce Manifold Representation Basis Learning (MARBLE),
a fully unsupervised representation-learning framework for non-linear dynamical systems.
Our approach combines empirical dynamical modelling and geometric deep learning to
transform neural activations during a set of trials into statistical distributions of
local flow fields (LFFs). Our central insight is that LFFs vary continuously over the
neural manifold, allowing for unsupervised learning, and are preserved under different
manifold embeddings, allowing the comparison of neural computations across networks and
animals.
We show that MARBLE offers a well-defined similarity metric between different neural
systems that is expressive enough to compare computations and detect fine-grained changes
in dynamics due to task variables, e.g., decision thresholds and gain modulation. Being
unsupervised, MARBLE is uniquely suited to biological discovery. Indeed, we show that it
discovers more interpretable neural representations in several motor, navigation and
cognitive tasks than generative models such as LFADS or (semi-)supervised models such as
CEBRA. Intriguingly, this interpretability implies significantly higher decoding
performance than state-of-the-art. Our results suggest that using the manifold structure
yields a new class of algorithms with higher performance and the ability to assimilate
data across experiments.
Time: 17:00, Wednesday, March 20., 2024.
Location: CEU Budapest site (1051 Budapest, Nádor u. 15.) N13. building, room 302.*
and Zoom (Meeting ID: 975 2152
9826<https://ceu-edu.zoom.us/j/97521529826?pwd=bXhOTDNzK054VFd2cUdMcVVCMkdUUT09>
Passcode: 996748)
*Anyone not affiliated with CEU wishing to attend in-person in Budapest must RSVP to
vargai(a)ceu.edu to get access to the lecture hall.
Please, be informed that video/photo recording might take place at the event and the
edited version of the video material might be published to communicate or promote
CEU's activities. Please, find our Privacy Notice
here<https://www.ceu.edu/privacy>cy>.
Best regards,
Ildikó Varga
Department Coordinator (Budapest)
Department of Cognitive Science
[cid:68fe031a-e7d8-4e68-9418-d22496979ad0]
H-1051 Budapest
Nádor u. 15. FT room 404.
tel: +36-1 327-3000 2941
http://www.ceu.edu<http://www.ceu.edu/>
http://cognitivescience.ceu.edu<http://cognitivescience.ceu.edu/>
______________________________________________
Subscribe by sending an empty mail to talks-subscribe(a)cogsci.ceu.edu
Unsubscribe by sending an empty mail to talks-unsubscribe(a)cogsci.ceu.edu
