Entropy-Based Decoding for Retrieval-Augmented Large Language Models

2024-06-25

arXiv.org

AI Summary

To address the decline in factual accuracy of retrieval-augmented large language models (RAG) caused by noisy external documents and interference from internal parametric knowledge, this paper proposes a training-free, entropy-guided decoding framework. The method introduces two novel mechanisms: (1) document-parallel entropy ensemble decoding, which samples multiple document paths in parallel and performs weighted ensemble of their outputs based on per-path output entropy to suppress low-confidence responses; and (2) cross-layer contrastive decoding, which dynamically identifies and attenuates unreliable, high-entropy internal knowledge activations by measuring KL divergence between hidden state distributions across Transformer layers. Evaluated on multiple open-domain question-answering benchmarks, the approach significantly improves answer factuality and effectively mitigates model “distractibility” without requiring fine-tuning or introducing additional parameters.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Augmenting Large Language Models (LLMs) with retrieved external knowledge has proven effective for improving the factual accuracy of generated responses. Despite their success, retrieval-augmented LLMs still face the distractibility issue, where the generated responses are negatively influenced by noise from both external and internal knowledge sources. In this paper, we introduce a novel, training-free decoding method guided by entropy considerations to mitigate this issue. Our approach utilizes entropy-based document-parallel ensemble decoding to prioritize low-entropy distributions from retrieved documents, thereby enhancing the extraction of relevant information of context. Additionally, it incorporates a contrastive decoding mechanism that contrasts the obtained low-entropy ensemble distribution with the high-entropy distribution derived from the model's internal knowledge across layers, which ensures a greater emphasis on reliable external information. Extensive experiments on open-domain question answering datasets demonstrate the superiority of our method.

Problem

• Mitigate distractibility in retrieval-augmented LLMs
• Enhance relevant information extraction
• Prioritize low-entropy distributions for accuracy

Innovation

• Entropy-based decoding method
• Document-parallel ensemble decoding
• Contrastive decoding mechanism