On the Computational Tractability of Robust Bandits
The paper pins robust bandits’ efficient-learning case to a narrow tractable boundary.
Kosoy and Pathak identify a special case of robust bandits with a polynomial-time learner and near-\(\sqrt{T}\) regret. They also show that several slight extensions become NP-hard, suggesting the tractable setup is close to the limit. The work frames this as progress on computationally efficient unrealizable learning, with AI alignment cited as a motivating application. ArXiv · AI/CL/LG's note
Kosoy and Pathak identify a special case of robust bandits with a polynomial-time learner and near-\(\sqrt{T}\) regret. They also show that several slight extensions become NP-hard, suggesting the tractable setup is close to the limit. The work frames this as progress on computationally efficient unrealizable learning, with AI alignment cited as a motivating application. ArXiv · AI/CL/LG's note
score 4