Evgenii Nikishin

I am a researcher at OpenAI, working on improving the capabilities of language models to solve scientific problems.

I received a PhD in Computer Science from Mila, University of Montreal advised by Pierre-Luc Bacon and Aaron Courville. Before that, I studied at Cornell University, Higher School of Economics, and Lomonosov Moscow State University. During my studies, I interned with the RL team at DeepMind and at ETH Zürich.

Email  /  CV  /  Google Scholar  /  Twitter  /  GitHub

Research
Maxwell’s Demon at Work: Efficient Pruning by Leveraging Saturation of Neurons
Simon Dufort-Labbé, Pierluca D’Oro, Evgenii Nikishin, Razvan Pascanu, Pierre-Luc Bacon, Aristide Baratin
Arxiv
[PDF]

We show that neuron saturation, which traditionally was viewed as undesirable, could instead lead to sparse yet accurate networks.

The Curse of Diversity in Ensemble-Based Exploration
Zhixuan Lin, Pierluca D'Oro, Evgenii Nikishin, Aaron Courville
ICLR 2024
[PDF, Poster]

We demonstrate that individual ensemble members in RL exhibit surprisingly low performance (whilst aggregate returns are adequate) and propose a remedy.

Deep Reinforcement Learning with Plasticity Injection
Evgenii Nikishin, Junhyuk Oh, Georg Ostrovski, Clare Lyle, Razvan Pascanu, Will Dabney, André Barreto
NeurIPS 2023 (Spotlight); also ICLR 2023 Workshop Track (Spotlight)
[PDF, Poster]

We propose an intervention for diagnosing the loss of plasticity phenomenon in RL and dynamically growing neural networks in RL for increasing computational efficiency.

Understanding Plasticity in Neural Networks
Clare Lyle, Zeyu Zheng, Evgenii Nikishin, Bernardo Ávila Pires, Razvan Pascanu, Will Dabney
ICML 2023 (Oral)
[PDF]

An analysis of plasticity loss showing its relation to pathological loss landscapes and demonstrating efficiency of layer normalization to mitigate it.

Sample-Efficient Reinforcement Learning by Breaking the Replay Ratio Barrier
Pierluca D'Oro*, Max Schwarzer*, Evgenii Nikishin, Pierre-Luc Bacon, Marc G. Bellemare, Aaron Courville
ICLR 2023 (Oral); also NeurIPS 2022 Workshop Track
[PDF, Poster, Code]

Resets unlock increasing sample efficiency by scaling the number of updates per environment step.

The Primacy Bias in Deep Reinforcement Learning
Evgenii Nikishin*, Max Schwarzer*, Pierluca D'Oro*, Pierre-Luc Bacon, Aaron Courville
ICML 2022; also RLDM 2022
[PDF, Short RLDM version, Code, Poster, Blog]

We identify a damaging tendency of deep RL agents to overfit early experiences and propose a simple yet powerful remedy based on periodic resetting of a part of the agent.

Control-Oriented Model-Based Reinforcement Learning with Implicit Differentiation
Evgenii Nikishin, Romina Abachi, Rishabh Agarwal, Pierre-Luc Bacon
AAAI 2022; also ICML 2021 Workshop Track
[PDF, Code, Poster]

A model learning method for RL that directly optimizes the sum of rewards instead of likelihood, a proxy to the agent's objective.

Quantifying and Understanding Adversarial Examples in Discrete Input Spaces
Volodymyr Kuleshov, Evgenii Nikishin, Shantanu Thakoor, Tingfung Lau, Stefano Ermon
ArXiv
[PDF]

We propose an algorithm to construct discrete token adversarial examples based on the notion of synonyms.

Unsupervised Domain Adaptation with Shared Latent Dynamics for Reinforcement Learning
Evgenii Nikishin, Arsenii Ashukha, Dmitry Vetrov
NeurIPS 2019 Workshop Track
[PDF, Code, Poster]

Domain adaptation via learning shared dynamics in a latent space with adversarial matching of latent states.

Improving Stability in Deep Reinforcement Learning with Weight Averaging
Evgenii Nikishin, Pavel Izmailov, Ben Athiwaratkun, Dmitrii Podoprikhin, Timur Garipov, Pavel Shvechikov, Dmitry Vetrov, Andrew Gordon Wilson
UAI 2018 Workshop Track
[PDF, Poster]

Averaging weights during training of an RL agent stabilizes the achieved performance.

Last update: Oct 2024.
Credit for the template to Jon Barron.