Research

I am interested in building adaptive agents that can safely navigate complex environments to achieve desired outcomes. To this end, I am focussed on the framework of Bayesian model-based reinforcement learning. I believe it provides the natural basis for building robust and adaptive artificial agents that can deal with stochastic and partially-observable environments.

More specifically, my research has been dedicated to creating better world models (latent-variable generative modelling) and model-based agents with Bayesian planning objectives.


Modelling non-reinforced preferences using selective attention
N Sajid, P Tigas, Z Fountas, Q Guo, A Zakharov, L Da Costa
NeurIPS 2022, WiML


Variational predictive routing with nested subjective timescales
A Zakharov, Q Guo, Z Fountas
ICLR 2022


Bayesian sense of time in biological and artificial brains
Z Fountas*, A Zakharov* (*equal contribution)
Book chapter in Time and Science 2022, World Scientific Publishing


Exploration and preference satisfaction trade-off in reward-free learning
N Sajid, P Tigas, A Zakharov, Z Fountas, K Friston
ICML 2021, WURL


Episodic memory for subjective-timescale models
A Zakharov, M Crosby, Z Fountas
ICML 2021, WURL


Geometric Deep Learning for Post-Menstrual Age Prediction
V Vosylius, A Wang, C Waters, A Zakharov, et al.
International Workshop on Graphs in Biomedical Image Analysis 2020