Search papers, labs, and topics across Lattice.
University of T眉bingen
3
0
8
By integrating diverse biochemical data through modality masking, Mol-JEPA outperforms traditional molecular models, setting a new standard for predictive accuracy in drug discovery.
Steering interventions reveal that embedding adjustments neutralize gender bias, while attention modifications can strategically shift it, highlighting the complexity of relevance signals in retrieval models.
Token-overlap and embedding-based metrics can't tell the difference between correct and contradictory statements, but MATCHA can, boosting accuracy by over 20% on TruthfulQA.
