
Oleg Smirnov
I’m a senior principal researcher at Microsoft Gaming and King
studio, where I focus on behavior modeling and alignment of AI agents
with human players. My broader research interests include geometric
methods (particularly deep learning in non-Euclidean domains and
Riemannian optimization), uncertainty quantification, and Bayesian
learning approaches.
Prior to joining Microsoft, I was a senior applied scientist at
Amazon and a member of technical staff at Spotify. My background is in
Applied Mathematics from Kyiv Polytechnic Institute, where I was advised
by Pavlo Maslyanko.
Outside of research, I enjoy applying gradient ascent in geographic
settings, also known as alpinism.
[Email] [Google
Scholar] [OpenReview]
[arXiv] [GitHub] [Substack]
Preprints
- O. Smirnov, S. Ennadir, J. Pertoft, B. Hjaltason, S. Karimi.
“Considering Context: When World Models Need Context Encoders.”
arXiv:2610.06651, 2026 [pdf]
- O. Smirnov, J. Pertoft, S. Karimi.
“When Agents Play Differently: Localizing Human–Agent Behavioral Divergence with Conditional Distributional Testing.”
arXiv:2611.26012, 2026 [pdf]
- S. Ennadir, E.M. Achour, O. Smirnov, L. Cao, E.H. Bergou.
“When Less Is More: Uncovering the Robustness Advantage of Model Pruning.”
2026 [pdf]
- O. Smirnov.
“tensorflow-riemopt: A Library for Optimization on Riemannian Manifolds.”
arXiv:2105.13921, 2021 [pdf]
Workshop papers
- O. Smirnov, S. Ennadir, J. Pertoft, B. Hjaltason, S. Karimi.
“When Does Dynamics Randomization Test Context Inference?”
In Do Robots Need World Models? workshop at CoRL,
2026 [pdf]
- O. Smirnov.
“The Language You Ask In: Language-Conditioned Ideological Divergence in LLM Analysis of Contested Political Documents.”
In Pluralistic Value Alignment of Large Language Models (PlurVA-LLM) workshop at AACL-IJCNLP,
2026 [pdf]
- S. Ennadir*, L. Zólyomi*, O. Smirnov*.
“Understanding Early Collapse in Predictive World-Model Pretraining.”
In World Models: Understanding, Modelling and Scaling workshop at ICLR,
2026 [pdf]
- F. Cornell*, O. Smirnov*, G. Gandler, L. Cao.
“Are We Really Measuring Progress? Transferring Insights from Evaluating Recommender Systems to Temporal Link Prediction.”
In Temporal Graph Learning (TGL) workshop at KDD,
2025 (Oral) [pdf]
- F. Cornell*, O. Smirnov*, G. Gandler, L. Cao.
“On the Power of Heuristics in Temporal Graphs.”
In I Can’t Believe It’s Not Better (ICBINB) workshop at ICLR,
2025 (Didactic Award nomination)
[pdf]
- F. Rietz, O. Smirnov, S. Karimi, L. Cao.
“Prompt-Tuning Bandits: Enabling Few-Shot Generalization for Efficient Multi-Task Offline RL.”
In Generalizing from Limited Resources in the Open World (GLOW) workshop at IJCAI,
2025 (Best Paper)
[pdf]
- T. Wang*, M. Honari-Jahromi*, S. Katsarou, O. Mikheeva, T. Panagiotakopoulos,
O. Smirnov, L. Cao, S. Asadi.
“Understanding Players as If They Are Talking to the Game in a Customized Language.”
In Customizable NLP (CustomNLP4U) workshop at EMNLP,
2024 [pdf]
- J. Talur, O. Smirnov, P. Missault.
“Few-Shot Out-of-Domain Intent Detection with Covariance-Corrected Mahalanobis Distance.”
In Uncertainty Reasoning and Quantification in Decision Making (UDM) workshop at AAAI,
2023 [pdf]
Conferences & journals
- L. Zólyomi, T. Wang, S. Ennadir, O. Smirnov, L. Cao.
“Towards Unified Approaches in Self-Supervised Event Stream Modeling: Progress and Prospects.”
In JAIR,
2026 [pdf]
- S. Ennadir*, O. Smirnov*, Y. Abbahaddou, L. Cao, J.F. Lutzeyer.
“Enhancing Graph Classification Robustness with Singular Pooling.”
In NeurIPS,
2025 [pdf]
- S. Ennadir*, L. Zólyomi*, O. Smirnov, T. Wang, J. Pertoft, F. Cornell,
L. Cao.
“Pool Me Wisely: On the Effect of Pooling in Transformer-Based Models.”
In NeurIPS,
2025 [pdf]
- F. Rietz, O. Smirnov, S. Karimi, L. Cao.
“Prompt Tuning Decision Transformers with Structured and Scalable Bandits.”
In NeurIPS,
2025 [pdf]
- S. Ennadir, Y. Jedra, O. Smirnov, L. Cao.
“Virtual Nodes Go Temporal.”
In Learning on Graphs (LoG),
2025 (Oral) [pdf]
- S. Ennadir*, G. Gandler*, F. Cornell*, L. Cao, O. Smirnov, T. Wang,
L. Zólyomi, B. Brinne, S. Asadi.
“Expressivity of Representation Learning on Continuous-Time Dynamic Graphs: An Information-Flow Centric Review.”
In TMLR,
2025 (Survey Certification)
[pdf]
Projects
Mountaineering
Teaching