TTT-Embed：测试时优化查询嵌入以最大化排名感知奖励
原标题：Test-Time Optimization of Query Embeddings with Ranking Aware Reward Maximization
AI 摘要
arXiv 论文提出 TTT-Embed 框架，在冻结嵌入模型的输出空间中学习轻量级残差向量，以蒸馏重排序器或 LLM 的排名奖励，无需访问模型权重或修改索引。实验覆盖 5 个嵌入模型和 15 个 MTEB 检索任务，TTT-Embed 将测试时检索性能提升最高 +8.36 nDCG@10，并能泛化到未见查询和任务，同时通过保持基础权重冻结来缓解灾难性遗忘。该框架支持全局、任务级和查询级三种共享范围，在固定奖励预算下实现可复用性与特异性的权衡。
正文节选
Test-Time Optimization of Query Embeddings with Ranking Aware Reward Maximization Abstract Dense retrievers rank documents using vector similarity between a frozen encoder and a precomputed index. While test-time ranking rewards from a reranker or LLM judge can improve results, existing methods discard this signal after a single query. Updating the retriever’s weights makes rewards reusable, but this requires parameter access—which is unavailable for closed-source models—and is computationally p