Webinar Review: Emergent Computation and Learning from Assemblies of Neurons

Our trainees review webinars in their given fields and share abstracts to help colleagues outside their discipline make an informed choice about watching them. As our program bridges diverse disciplines, these abstracts are beneficial for our own group in helping one another gain key knowledge in each other’s fields. We are happy to share these here for anyone else who may find them helpful.

Emergent Computation and Learning from Assemblies of Neurons

Part of Georgia Tech’s 2022-2023 Neuro Seminar Series

Dr. Santosh Vempala, College of Computing, Georgia Institute of Technology

September 19, 2022

Watch on Georgia Tech’s website >>

Ryan MillerSummary and Analysis by Ryan Miller:

In this talk, Dr. Vempala starts his motivation of the talk with raising the question “how does the mind emerge from the brain?” Where, despite accelerating progress in neuroscience and increasing insight in cognitive science, an overarching theory remains elusive to this question. Therefore, a computational system consistent with the current understanding of the brain that explains cognitive phenomena can be useful to propose a theory to this question, and as such. assembly calculus was proposed.

Assembly calculus is a formal probabilistic model of the brain that takes in one basic data type, computes few elementary operations, and is characterized as a completeness theorem. The brain model consists of a finite number of brain regions each containing neurons and displays an inhibition mechanism. From this model a key observation arises, which is that brains learn from assemblies, and occurs locally though plasticity.

With this observation, Dr. Vempala maps mild learning through assemblies to real-world examples of learning through assembly of neural firing frequencies. This illustration articulates how the observed patterns of learning in simplified computational models can be used as foundational piece for theorizing learning pathways and eventual understanding of cognition through neural assemblies. 

Overall, Dr. Vempala does an elegant job of describing the framework of the computational model to an audience broad enough to span to fields outside of computing. This is achieved through a robust description of defining parameters of the system followed by examples with increasing complexity. I found the work to be strongly motivated and limitations properly discussed. At the end, Dr. Vempala points out a few open problems related to assemblies and convergence, of which include: phase transitions in support size, possibly (non vanishing probability) limiting behaviors, and how to define assemblies as distributions.