Parnian Kassraie
Research Scientist, Google DeepMind
pkassraie [AT] google.com
Research
Ph.D. Research: This body of work focuses on intelligent agents capable of making discoveries in domains where objectives are ambiguous, and data is scarce. Instead of relying on fixed, expert-defined representation of such environments, we study how agents can curate their own data and learn useful representations of the problem space while solving it. We develop a theoretical framework for optimization under unknown representations and proposed two complementary approaches: [LIN], which offers efficient and provably optimal strategies in linear problem settings, and [NN], which leverages the flexibility of neural networks to handle more complex structures, e.g. graph domains. Together, these methods shed light on how we can move from trial-and-error exploration toward autonomous systems capable of genuine discovery.
LITE: Efficiently Estimating Gaussian Probability of Maximality
Nicolas Menet, Jonas Hübotter, Parnian Kassraie, Andreas Krause
In International Conference on Artificial Intelligence and Statistics (AISTATS), 2025
Bandits with Preference Feedback: A Stackelberg Game Perspective
Barna Pásztor*, Parnian Kassraie*, Andreas Krause
In Conference on Neural Information Processing Systems (NeurIPS), 2024
Progressive Entropic Optimal Transport Solvers
Parnian Kassraie, Aram-Alexandre Pooladian, Michal Klein, James Thornton, Jonathan Niles-Weed, Marco Cuturi
In Conference on Neural Information Processing Systems (NeurIPS), 2024
Instance-Dependent Generalization Bounds via Optimal Transport
Songyan Hou*, Parnian Kassraie*, Anastasis Kratsios*, Jonas Rothfuss*, Andreas Krause
Journal of Machine Learning Research, 2023
Anytime Model Selection in Linear Bandits
Parnian Kassraie, Nicolas Emmenegger, Andreas Krause, Aldo Pacchiano
In Conference on Neural Information Processing Systems (NeurIPS), 2023
Also in PAC-Bayes Meets Interactive Learning Workshop at ICML, 2023 (Oral)
Model-Based Optimization over Large Molecular Spaces
Miles Wang-Henderson*, Bartu Soyuer*, Parnian Kassraie, Andreas Krause, Ilija Bogunovic
In ICML Workshop on Structured Probabilistic Inference & Generative Modeling, 2023
Also In NeurIPS Workshop on AI for Drug Discovery, 2023
Hallucinated Adversarial Control for Conservative Offline Policy Evaluation
Jonas Rothfuss*, Bhavya Sukhija*, Tobias Birchler*, Parnian Kassraie, Andreas Krause
In Conference on Uncertainty in Artificial Intelligence (UAI), 2023
Lifelong Bandit Optimization: No prior and No Regret
Felix Schur*, Parnian Kassraie*, Jonas Rothfuss, Andreas Krause
In Conference on Uncertainty in Artificial Intelligence (UAI), 2023
Graph Neural Network Bandits
Parnian Kassraie, Andreas Krause, Ilija Bogunovic
In Conference on Neural Information Processing Systems (NeurIPS), 2022
Meta-Learning Hypothesis Spaces for Sequential Decision-making
Parnian Kassraie, Jonas Rothfuss, Andreas Krause
In International Conference on Machine Learning (ICML), 2022
Also in Safe, Anytime-Valid Inference workshop (SAVI), 2022
Also in Adaptive Experimental Design and Active Learning Workshop (ReALML), 2022
Neural Contextual Bandits without Regret
Parnian Kassraie, Andreas Krause
In International Conference on Artificial Intelligence and Statistics (AISTATS), 2022
Anytime Model Selection in Linear Bandits
Parnian Kassraie, Nicolas Emmenegger, Andreas Krause, Aldo Pacchiano
In Conference on Neural Information Processing Systems (NeurIPS), 2023
Also in PAC-Bayes Meets Interactive Learning Workshop at ICML, 2023 (Oral)
Lifelong Bandit Optimization: No prior and No Regret
Felix Schur*, Parnian Kassraie*, Jonas Rothfuss, Andreas Krause
In Conference on Uncertainty in Artificial Intelligence (UAI), 2023
Meta-Learning Hypothesis Spaces for Sequential Decision-making
Parnian Kassraie, Jonas Rothfuss, Andreas Krause
In International Conference on Machine Learning (ICML), 2022
Also in Safe, Anytime-Valid Inference workshop (SAVI), 2022
Also in Adaptive Experimental Design and Active Learning Workshop (ReALML), 2022
Model-Based Optimization over Large Molecular Spaces
Miles Wang-Henderson*, Bartu Soyuer*, Parnian Kassraie, Andreas Krause, Ilija Bogunovic
In ICML Workshop on Structured Probabilistic Inference & Generative Modeling, 2023
Also In NeurIPS Workshop on AI for Drug Discovery, 2023
Hallucinated Adversarial Control for Conservative Offline Policy Evaluation
Jonas Rothfuss*, Bhavya Sukhija*, Tobias Birchler*, Parnian Kassraie, Andreas Krause
In Conference on Uncertainty in Artificial Intelligence (UAI), 2023
Graph Neural Network Bandits
Parnian Kassraie, Andreas Krause, Ilija Bogunovic
In Conference on Neural Information Processing Systems (NeurIPS), 2022
Neural Contextual Bandits without Regret
Parnian Kassraie, Andreas Krause
In International Conference on Artificial Intelligence and Statistics (AISTATS), 2022
Progressive Entropic Optimal Transport Solvers
Parnian Kassraie, Aram-Alexandre Pooladian, Michal Klein, James Thornton, Jonathan Niles-Weed, Marco Cuturi
In Conference on Neural Information Processing Systems (NeurIPS), 2024
Instance-Dependent Generalization Bounds via Optimal Transport
Songyan Hou*, Parnian Kassraie*, Anastasis Kratsios*, Jonas Rothfuss*, Andreas Krause
Journal of Machine Learning Research, 2023