Osbert Bastani

Unlock continuous robot learning with Tether, the first method to achieve hours of autonomous multi-task play in the real world, starting from just a handful of demonstrations.

Forget Bayesian priors: swap regret bounds alone can align incentives for exploration in dynamic bandit settings with diverse agent beliefs.

Provably optimal program synthesis, guided by abstract interpretation, can now outperform existing methods in data classification DSLs.

Automating the creation of trajectory queries from examples slashes development time by eliminating manual parameter tuning.