Paper
Xiaohongshu et al. Propose UniNote: A Unified Multimodal Embedding Model
Xiaohongshu, Shanghai Jiao Tong University, Huazhong University of Science and Technology, and Beijing Institute of Technology jointly propose UniNote, which integrates multimodal representation learning and ranking capabilities into a single model. Through two-stage training (contrastive SFT for representation enhancement + reinforcement learning for ranking optimization), UniNote achieves performance comparable to the two-stage retrieval-ranking paradigm in a single embedding pass, significantly reducing latency.
Read the original (opens in a new tab)
News stream data aggregated by AI HOT