TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings
#TSEmbed #multimodal embeddings #task scaling #universal embeddings #AI adaptation
📌 Key Takeaways
- TSEmbed is a new method for scaling tasks in universal multimodal embeddings.
- It enhances the ability of embeddings to handle diverse tasks without retraining.
- The approach improves performance across multiple modalities like text and images.
- TSEmbrid demonstrates potential for more efficient AI model adaptation.
📖 Full Retelling
arXiv:2603.04772v1 Announce Type: cross
Abstract: Despite the exceptional reasoning capabilities of Multimodal Large Language Models (MLLMs), their adaptation into universal embedding models is significantly impeded by task conflict. To address this, we propose TSEmbed, a universal multimodal embedding framework that synergizes Mixture-of-Experts (MoE) with Low-Rank Adaptation (LoRA) to explicitly disentangle conflicting task objectives. Moreover, we introduce Expert-Aware Negative Sampling (EA
🏷️ Themes
AI Research, Multimodal Learning
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
--> Computer Science > Computation and Language arXiv:2603.04772 [Submitted on 5 Mar 2026] Title: TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings Authors: Yebo Wu , Feng Liu , Ziwei Xie , Zhiyuan Liu , Changwang Zhang , Jun Wang , Li Li View a PDF of the paper titled TSEmbed: Unlocking Task Scaling in Universal Multimodal Embeddings, by Yebo Wu and 6 other authors View PDF HTML Abstract: Despite the exceptional reasoning capabilities of Multimodal Large Language Models , their adaptation into universal embedding models is significantly impeded by task conflict. To address this, we propose TSEmbed, a universal multimodal embedding framework that synergizes Mixture-of-Experts with Low-Rank Adaptation to explicitly disentangle conflicting task objectives. Moreover, we introduce Expert-Aware Negative Sampling , a novel strategy that leverages expert routing distributions as an intrinsic proxy for semantic similarity. By dynamically prioritizing informative hard negatives that share expert activation patterns with the query, EANS effectively sharpens the model's discriminative power and refines embedding boundaries. To ensure training stability, we further devise a two-stage learning paradigm that solidifies expert specialization before optimizing representations via EANS. TSEmbed achieves state-of-the-art performance on both the Massive Multimodal Embedding Benchmark and real-world industrial production datasets, laying a foundation for task-level scaling in universal multimodal embeddings. Subjects: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2603.04772 [cs.CL] (or arXiv:2603.04772v1 [cs.CL] for this version) https://doi.org/10.48550/arXiv.2603.04772 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Yebo Wu [ view email ] [v1] Thu, 5 Mar 2026 03:43:52 UTC (6,038 KB) Full-text links: Access Paper: View a PDF of the paper titled TSEmbed: Unlocking Task Scaling in Univ...
Read full article at source