LLM-based dense retrievers generalize better when instruction-tuned but pay a specialization tax when optimized for reasoning; they resist typos and corpus poisoning better than encoder-only baselines yet remain vulnerable to semantic perturbations, with larger models and certain embedding geometry,
In: Proceedings of the 45th Interna- tional ACM SIGIR Conference on Research and Development in Informa- tion Retrieval
7 Pith papers cite this work, alongside 35 external citations. Polarity classification is still indexing.
7
Pith papers citing it
35
external citations · external index
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
HieraRAG shows optimal RAG benchmark granularity varies by dimension, with complexity favoring fine-grained categories and a new Coherence Ratio measuring category structure.
A dual hierarchical RL framework with two agents coordinates high-level dialogue strategy and low-level question generation to emulate judicial questioning and extract key information from Supreme Court arguments, outperforming baselines.
MemoryBench shows that state-of-the-art LLM memory systems do not reliably learn from simulated user feedback and are often outperformed by naive RAG.
DMICF models interactions from user- and item-centric perspectives with a macro-micro prototype-aware variational encoder and dimension-wise intent alignment to improve collaborative filtering.
Small language models can run RAG generation on-device without GPUs in reasonable time.
MetaOrch is a deep learning orchestrator that selects optimal agents in multi-agent systems via fuzzy-scored supervision labels, reporting 86.3% accuracy in simulations over random and round-robin baselines.
citing papers explorer
-
On the Robustness of LLM-Based Dense Retrievers: A Systematic Analysis of Generalizability and Stability
LLM-based dense retrievers generalize better when instruction-tuned but pay a specialization tax when optimized for reasoning; they resist typos and corpus poisoning better than encoder-only baselines yet remain vulnerable to semantic perturbations, with larger models and certain embedding geometry,
-
How Fine-Grained Should a RAG Benchmark Be? A Hierarchical Framework for Synthetic Question Generation
HieraRAG shows optimal RAG benchmark granularity varies by dimension, with complexity favoring fine-grained categories and a new Coherence Ratio measuring category structure.
-
Dual Hierarchical Dialogue Policy Learning for Legal Inquisitive Conversational Agents
A dual hierarchical RL framework with two agents coordinates high-level dialogue strategy and low-level question generation to emulate judicial questioning and extract key information from Supreme Court arguments, outperforming baselines.
-
MemoryBench: A Benchmark for Memory and Continual Learning in LLM Systems
MemoryBench shows that state-of-the-art LLM memory systems do not reliably learn from simulated user feedback and are often outperformed by naive RAG.
-
Dual-Perspective Disentangled Multi-Intent Alignment for Enhanced Collaborative Filtering
DMICF models interactions from user- and item-centric perspectives with a macro-micro prototype-aware variational encoder and dimension-wise intent alignment to improve collaborative filtering.
-
Little Brains, Big Feats: Exploring Compact Language Models
Small language models can run RAG generation on-device without GPUs in reasonable time.
-
Neural Orchestration for Multi-Agent Systems: A Deep Learning Framework for Optimal Agent Selection in Multi-Domain Task Environments
MetaOrch is a deep learning orchestrator that selects optimal agents in multi-agent systems via fuzzy-scored supervision labels, reporting 86.3% accuracy in simulations over random and round-robin baselines.