Bahti Zakirov

I'm a PhD student at ISTA in Vienna, supervised by Gašper Tkačik. Previously I've spent some glorious years at the Francis Crick Institute in London, and at CUNY in New York.

Email  /  Scholar    

Research

I’m working on characterizing the statistical computations that biological systems (brains, cells, otherwise) perform to autonomously discover and adapt to structure in the world from finite experiences. I’m also interested in developing artificial systems that can do the same- aiming to automate some aspects of the scientific process. I approach both of these directions using tools from information theory and probabilistic machine learning.

Selected Projects

We investigate a normative role for episodic replay in systems consolidation in i) enabling routing experiences for consolidation to appropriate schemas and ii) shaping schemas' inductive biases. This extends complementary learning systems views of Hippocampal-Neocortical memory consolidation.

We developed a Bayesian meta-learning framework to construct flexible probabilistic models representing structure from neuro-theory knowledge. These theory-surrogate models resulted in both an inductive bias to improve data-driven predictive fits on real neural data, and a more automated way to validate and update the theory itself from the data using exact Bayesian model comparison.

We studied how endothelial cells use their embodiment and sensorimotor coordination to add informational structure to inputs from their local chemical environment (i.e. "Active Perception") in order to facilitate adaptive behaviour. This active perception speeded up Notch-mediated tip-cell selection, the timing of which is crucial for blood vessel morphology.

Thanks for the website, Jon Barron!