Sattar Vakili

Principal AI Research Manager

MediaTek Research

Visiting Researcher · Wellcome Sanger Institute
PhD (ECE) · Cornell University

Research Interests

Sequential Decision-Making (RL, Bandits and Bayesian Optimisation) · Foundation and Generative Models · AI for Science and Intelligent Systems

Latest

Jan 2026: A Finite Time Analysis of Thompson Sampling for Bayesian Optimization with Preferential Feedback is accepted at AISTATS 2026.
Jan 2026: Reinforcement Learning Using Known Invariances is accepted at AISTATS 2026.
Oct 2025: Invited talk on Decision-Making Under Uncertainty: AI with Human-in-the-Loop Perspective at the Workshop on Causal AI in Healthcare Policy & Practice, Oxford University.
Jun 2025: Reinforcement Learning with Thompson Sampling: No-Regret Performance over Finite Horizons is accepted at the ICML 2025 workshop Exploration in AI Today.
May 2025: Will be giving a tutorial on Foundation Models for Communication Systems at IEEE GLOBECOM, December 2025, Taipei.
Mar 2025: Giving a Tech Talk on AI and Communication at Cambridge University, Mar 10th.
Feb 2025: Giving a presentation at the Computational Statistics and Machine Learning (OxCSML) seminar series, Oxford University, Feb 21st.
Dec 2024: Presenting our NeurIPS paper at local meetup, Cambridge University, Dec 6th.
Oct 2024: Trieste: Efficiently Exploring The Depths of Black-box Functions with TensorFlow is accepted at NeurIPS 2024 Workshop on Bayesian Decision-making and Uncertainty.
Sep 2024: Adversarial Contextual Bandits Go Kernelized will be presented at European Workshop on Reinforcement Learning (EWRL 2024).
Sep 2024: Kernel-Based Function Approximation for Average Reward Reinforcement Learning will be presented at European Workshop on Reinforcement Learning (EWRL 2024).
Jul 2024: Giving a tutorial on Recent Advances of Statistical Reinforcement Learning at UAI 2024, July 15th in Barcelona. [slides]
Jun 2024: Involved in organizing a local ICML meetup on July 12th in London.
Jun 2024: Reward-Free Kernel-Based Reinforcement Learning is accepted at ICML 2024.
Apr 2024: Giving a tutorial on Recent Advances of Statistical Reinforcement Learning at UAI 2024, July 15th in Barcelona.
Dec 2023: Optimal Regret Bounds for Collaborative Learning in Bandits is accepted at Algorithmic Learning Theory (ALT) 2024.
Dec 2023: Adversarial Contextual Bandits Go Kernelized is accepted at Algorithmic Learning Theory (ALT) 2024.
Dec 2023: Presenting our NeurIPS paper at local meetup, Cambridge University, Dec 8th.
Earlier news (2021–2023)
Oct 2023: Collaborative Learning in Kernel-based Bandits for Distributed Users is accepted at IEEE Transactions on Signal Processing.
Oct 2023: Giving a talk at the Inria Scool seminar series at University of Lille.
Oct 2023: Adversarial Contextual Bandits Go Kernelized is available on arXiv.
Aug 2023: Check out Tor Lattimore's response to the open problem on online confidence intervals for RKHS elements.
Jul 2023: Giving an online lecture at FeDucation seminar series (Florida International University).
Jun 2023: Giving a seminar on kernel-based reinforcement learning at Deepmind/Ellis CSML seminar series.
May 2023: Giving an invited talk on kernel-based RL at the London Symposium on Information Theory.
Apr 2023: Delayed Feedback in Kernel Bandits is accepted at ICML 2023.
Apr 2023: Image generation with shortest path diffusion is accepted at ICML 2023.
Feb 2023: Delayed Feedback in Kernel Bandits is available on arXiv.
Jan 2023: Sample Complexity of Kernel-Based Q-Learning is accepted at AISTATS 2023.
Dec 2022: Presenting Gradient Descent: Robustness to Adversarial Corruption at OPT2022 workshop at NeurIPS 2022, New Orleans.
Oct 2022: Near-Optimal Collaborative Learning in Bandits has been designated as an Oral presentation at NeurIPS 2022.
Jul 2022: Presenting an open problem on noise-free kernel-based bandit at COLT 2022, London.
May 2022: Near-Optimal Collaborative Learning in Bandits is available on arXiv.
May 2022: Improved Convergence Rates for Sparse Approximation Methods in Kernel-Based Learning is accepted at ICML 2022 for a Spotlight presentation.
Oct 2021: Optimal Order Simple Regret for Gaussian Process Bandits is accepted at NeurIPS 2021.
Aug 2021: Moderating the "Bandits, RL and Control" session at COLT 2021.
Aug 2021: Presenting Tight Online Confidence Intervals for RKHS Elements at COLT 2021.