Publications

Utilizing an alignment-first strategy to transfer advanced reasoning skills to a non-reasoning model (NeurIPS).

An efficient solution to enforce the well-typedness of LLM functions. (EurIPS)

An RL proxy task that trains VLMs to localize synthetic hallucinations injected into human-written captions. (NeurIPS)

An evaluation architecture that incorporates reasoning and iterative refinement. (NeurIPS)

An uncertainty measure defined as the cross-entropy between predictive distribution and unknown true distribution. (NeurIPS)

An RL approach that leverages multiple rollouts across paraphrased set to assign group similarity rewards. (NeurIPS)

An efficient, scalable method for generating synthetic pre-training data to enable meta-learning of decision trees. (NeurIPS)

A unified end-to-end architecture that couples a sequence encoder with a GNN. (NeurIPS)

A routing approach that assigns each problem to the smallest model likely to solve it, reducing compute. (NeurIPS)