Yaniv Romano

Unlabeled data can significantly boost the power of hypothesis tests, even when predictions are noisy, offering a practical way to improve statistical inference in data-scarce regimes.

Reinforcement learning's Temporal Difference value estimation offers a surprisingly effective and theoretically grounded approach to calibrating uncertainty in vision-language-action models for robotics.

Unleashing the full potential of block diffusion drafting, DDTree's tree-based verification strategy dramatically accelerates speculative decoding by exploring multiple likely continuations in parallel.