Search papers, labs, and topics across Lattice.
Sorbonne Universit茅, CNRS, ISIR
2
0
4
0
SPLADE models can ditch their token-based vocabularies for a latent semantic space learned by Sparse Auto-Encoders, maintaining retrieval performance while boosting efficiency.
Forget bigger models: smarter training objectives like pairwise MarginMSE and listwise InfoNCE can boost cross-encoder performance as much as scaling the backbone architecture.
