Detailed walkthrough of our papers on multimodal learning, language models, and biomedical AI. Browse publications.
Filter articles by topic
4 of 4 articles
No articles match this topic.
A movie poster and its plot can tell us different things. We trace a prediction through I²MoE’s experts to see what each input contributes alone and what they reveal together.
Multimodal learning, Mixture of experts, Interpretability
Read articleA small rewrite can leave an answer unchanged while shifting a language model’s uncertainty. Scoring multiple paraphrases helps make prediction sets more stable—but the coverage guarantees still depend on specific assumptions.
Large language models, Uncertainty quantification, Conformal prediction
Read articleWith room for only a few examples in a prompt, each choice matters. UCS uses cluster frequencies to estimate what the selected demonstrations may still be missing.
Large language models, In-context learning, Coverage estimation
Read articleWill an antibody still synthesize successfully after its format changes? We bring sequence and structural context together to predict synthesis outcomes, then test the model on splits that reflect different experimental settings.
Multimodal learning, Biomedical AI, Antibody engineering
Read article