Researchers have developed MixLoRA-DSI, a new framework designed to efficiently update generative retrieval models with new documents without requiring full retraining. This method employs an expandable mixture of Low-Rank Adaptation experts and a layer-wise strategy that introduces new experts only when out-of-distribution documents are detected. Experiments on NQ320k and MS MARCO Passage datasets show that MixLoRA-DSI achieves better performance than full-model updates while incurring significantly lower training costs and minimal parameter overhead. AI
IMPACT This method could reduce the computational cost of updating large language models for information retrieval tasks.
RANK_REASON The cluster contains an academic paper detailing a new method for generative retrieval. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
