A zero-shot approach to fine-grained image captioning built on vision-language-action (VLA) models, producing detailed, object-level descriptions in a single pass while matching the quality of multi-pass pipelines at a fraction of their cost and latency.

Introduces reward-based φ-function state abstractions for the STAR estimator, enabling low-variance, low-bias off-policy evaluation that outperforms importance-sampling and model-based baselines across RL environments. Manuscript under submission; available upon request.

We introduce an optimal minority-veto strategy that is resilient to missing data and mitigates this bias to a large extent. For scenarios requiring even higher precision, we propose a novel regression-based framework that directly models the validator bias using a small set of human-annotated ground truth data. On a challenging code feedback task over 366 high-school Python programs, our regression approach reduces the maximum absolute error to just 1.2%, achieving a 2x improvement over the best-performing ensemble of 14 state-of-the-art LLMs.

Provide a bound on the cop number of graphs in terms of their 2-component order connectivity.

Paper shows that fixed-parameter tractability cannot be extended to capture all FO-definable problems. It answers this question by showing that FO model checking on induced subgraphs of hypercubes is as difficult as FO model checking on general graphs.