Scholar
Tor Lattimore
Google Scholar ID: fkDxJxcAAAAJ
Google DeepMind
Citations & Impact
Citations
8,099
H-index
36
i10-index
70
Publications
20
Co-authors
113
list available
Contact
Publications
Cited
10
Resume (English only)
Academic Achievements
- Completed a set of notes on bandit convex optimisation.
- The 'Convex Bandit' paper was accepted at the journal 'Mathematical Statistics and Learning'.
- Research results with Andras Gyorgy showed that the stability of mirror descent for some exploration distribution and estimators is upper bounded by the information ratio.
- Two partial monitoring papers were accepted at COLT.
- A preprint on bandit convex optimisation improved the dimension-dependence in the information-theoretic rate from d^9.5 to d^3 d^(2.5).
- Co-authored a paper with Shuai Li and Csaba on learning to rank in a linear model, accepted to ICML.
- Followup on the COLT paper on partial monitoring with Csaba, proposing a simple and efficient algorithm that matches the non-constructive upper bound.
Research Experience
- Research scientist at DeepMind, based in London.
- Completed a new book on bandits with Csaba, published by Cambridge University Press.
- Collaborated with Andras Gyorgy on connections between information-theoretic arguments and adversarial bandits based on mirror descent or follow the regularised leader.
- Co-authored a paper with Julian Zimmert showing that the information ratio is upper bounded by stability in certain linear settings.
- Worked with Johannes Kirschner and Andreas Krause on a linear version of partial monitoring.
- Co-authored a paper with Botao Hao on the asymptotics of linear contextual bandits.
- Collaborated with Csaba and Gellert on misspecified linear models in bandit and RL settings.
- Developed a new algorithm for tree search.
- Contributed to the bsuite RL proving ground.
- Conducted research on model selection for contextual bandits.
Background
- Research scientist, mainly working on algorithms for sequential decision making. Spent a lot of time in recent years on bandits.
Miscellany
- Personal interests include drawing posters using GoodNotes and Graphic software.
