Email: ivan.nazarov@skolkovotech.ru
Website: https://github.com/ivannz
Link to CV: https://github.com/ivannz/CV/blob/main/nazarov_eng.pdf
Level: PhD, Research Visit
Keywords: RL, ML, optimization, sparsification, hierarchical RL
Short bio / research interests / publications: I am a 35 year old with 12 year working experience. I have been deeply immersed in machine learning for the last seven years, during which worked on various project, both industrial and theoretically inclined, aimed at optimization and sparsification methods for DL.
I have successfully completed my PhD studies on “applied mathematics, computer science and engineering” at Skoltech in 2020. My PhD thesis on the topic of model sparsification: sparse-regularized matrix decompositions, variational dropout, and parameter pruning methods based on second-order loss approximation. I have one major paper at ICML2020 (https://proceedings.mlr.press/v119/nazarov20a.html) about variational Dropout for complex-valued deep networks.
In the last year i have been consumed by the exciting and challenging field of Reinforcement Learning and Optimal Control, specifically hierarchical policies, planning, MCTS and adaptive computations. My team took first place in the recent NeurIPS Nethack Challenge for our hybrid algorithmic-neural solution.