Search papers, labs, and topics across Lattice.
2
0
5
Standard multimodal fusion can hurt performance in emotion recognition, but this new approach knows when to drop modalities, leading to state-of-the-art results.
Training on semantically equivalent chart renderings in Python, R, and LaTeX unlocks surprisingly effective multi-lingual chart-to-code generation from a single model.
