Igor Kuznetsov

I’m a research engineer with a focus on deep reinforcement learning. My research interests are around memory mechanisms for intelligent agents, safety in reinforcement learning, and the connection between RL and language.

I obtained my bachelor's degree at ITMO University. Currently, I'm a senior engineer at Oxa, where I develop scenarios for autonomous solutions in simulation. Before that, I worked at JetBrains Research and Insilico Medicine.

cv | scholar | github | twitter

Research

Safer Reinforcement Learning by Going Off-policy: a Benchmark
Igor Kuznetsov
ICML, 2024 - NextGenAISafety Workshop
Paper | GitHub
A benchmark of off-polcy RL algorithms applied to SafeRL problems.
Guided Exploration in Reinforcement Learning via Monte Carlo Critic Optimization
Igor Kuznetsov
AAMAS, 2024 / ICML, 2022 - DARL Workshop
arxiv | GitHub
A directed exploration method based on exploiting a directional controller that is represented as an ensemble of Monte Carlo critics.
Solving Continuous Control with Episodic Memory
Igor Kuznetsov, Andrey Filchenkov
IJCAI, 2021
arxiv | poster | GitHub | talk
An off-policy model-free algorithm that exploits episodic memory to solve continuous control tasks.
Conditioning of Reinforcement Learning Agents and its Policy Regularization Application
Arip Asadulaev, Igor Kuznetsov, Gideon Stein, Andrey Filchenkov
ICML, 2020 - BIG Workshop
arxiv | talk | IEEE Access
The study of the outcome of Jacobian singular values regularization with application to reinforcement learning domain.