Hexuan (Matt) Wang

I'm broadly interested in natural language processing, multimodal learning, and machine learning. My focus is faithfulness: a model should be faithful to its own knowledge and behavior, doing what it knows and what it says it will do, and stay grounded in its sources, with each claim resting on evidence it can point to. It should also be verifiable: a person can check all of this, and check it efficiently. I work on this in cited text, over structured data, under temporal shift, and across modalities.