Georgia Institute of Technology
A research on disentangled neural signal dimensionality reduction: balancing group coupling and independence
Abstract
dc:description.abstractPoisson-distributed latent variable models are widely used in computational neuroscience, but differentiating through discrete stochastic samples remains challenging. Two approaches address this: Exponential Arrival Time simulation and Gumbel-SoftMax relaxation. We provide the comparison of these methods, along with practical guidance for practitioners. % Our main technical contribution is a modification to the EAT method that theoretically guarantees an unbiased first moment (exactly matching the firing rate), and reduces second-moment bias. We evaluate these methods on partially observable generalized linear models, where latent neural connectivity can be inferred from observed spike trains. % Our modified EAT method exhibits better overall performance, and substantially higher robustness to hyperparameter choices. Together, our results demonstrate the trade-offs between these methods and offer recommendations for practitioners working with Poisson latent variable models.
Degree
thesis:*- Name thesis:degree_name
- Robotics, MS
- Department thesis:degree_department
- Electrical and Computer Engineering
- Grantor
- Georgia Institute of Technology
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Zhao, Hanqi
- Advisor dc:contributor.advisor
-
- Wu, Anqi
- Committee members dc:contributor.committeemember
-
- Zhao, Ye
- Sponberg, Simon
Rights
- Language dc:language.iso
- English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1853/81634