Publications

A setting in which routers are trained on generated queries and answers produced from high-level task descriptions. (ACL)

A community-driven, human-annotated LID benchmark for the web domain, covering 109 languages. (ACL)

A template-first benchmark that factorizes reasoning type and cultural aspect across question languages. (ACL)

A massive-scale benchmark for evaluating retrieval-augmented VQA across languages and modalities. (CVPR)

A novel cultural adversarial robustness suite designed to assess VLMs’ stability against perturbed geographical cues. (CVPR)

A suite of dynamic guardian models offering novel flexibility by evaluating text based on user-defined policies. (ICLR)

A DPO that targets the most problematic parts of an output by assigning different preference weights. (ICLR)

A framework for multivariate time series forecasting using tabular foundation models. (ICLR)

A private text generation method that steers LLM generation using dataset vectors. (ICLR)