Search papers, labs, and topics across Lattice.
This paper introduces ECI_{sem}, a novel approach for selecting hard-negative sources in dense retrieval by leveraging frozen target-encoder embeddings to rank candidates without requiring fine-tuning. The method constructs a weighted residual information matrix that incorporates various semantic and lexical factors, achieving superior ranking of LLM negatives and hybrid sources on the MS MARCO dataset. Notably, ECI_{sem} aligns closely with the best BEIR transfer results across multiple models, demonstrating its effectiveness and robustness across various conditions.
ECI_{sem} ranks LLM negatives highest among non-hybrid sources, revealing that effective hard-negative selection can be achieved without fine-tuning.
Hard-negative source selection for dense retrieval is usually decided only after fine-tuning and downstream evaluation. We propose ECI_{sem}, a semantic residual variant of Effective Contrastive Information (ECI) that ranks candidate negative sources using frozen target-encoder embeddings. ECI_{sem} is training-free, not label-free: each scored example requires a query, a labeled positive, and an explicit candidate negative. ECI_{sem} builds a weighted residual information matrix from target consistency, semantic locality, lexical residuality, and a log-determinant diversity objective. On MS MARCO negative sources, in-family ECI_{sem} ranks LLM negatives highest among non-hybrid sources and Dense+LLM highest among hybrid sources, matching the strongest aggregate BEIR transfer results across DistilBERT, E5-base, and Contriever. Controlled ablations show that this alignment depends on using the target encoder family, while additional ablations show stability under sample-size, temperature, tokenizer, and IDF-corpus perturbations. The theory gives a local linearized link to loss reduction, while the empirical study treats downstream evaluation as the final test.
