Enhancing elusive clues in knowledge learning by contrasting attention of language models

2024-09-26

arXiv.org

AI Summary

To address low knowledge acquisition efficiency in language models trained on small-scale, knowledge-intensive corpora—caused by challenges in modeling long-range dependencies, overfitting to sparse co-occurrence patterns, and dispersion of informative cues—this paper proposes an attention-divergence-driven data augmentation method. It is the first to leverage cross-scale attention distribution discrepancies between large and small models during pretraining to automatically identify discriminative yet easily overlooked implicit knowledge cues. Based on these cues, we design a cue-guided token-dropout augmentation strategy that enhances training signal quality in a fully unsupervised manner. The method requires no additional annotations or hand-crafted rules. Experimental results demonstrate significant improvements in factual memory accuracy, validating that attention-behavior divergence serves as a transferable and scalable optimization signal for knowledge learning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Data Mining & Knowledge Management: Knowledge Acquisition from the Web

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

Causal language models acquire vast amount of knowledge from general text corpus during pretraining, but the efficiency of knowledge learning is known to be unsatisfactory, especially when learning from knowledge-dense and small-sized corpora. The deficiency can come from long-distance dependencies which are hard to capture by language models, and overfitting to co-occurrence patterns and distracting clues in the training text. To address these issues, the paper proposes a method to enhance knowledge learning during language model pretraining, by enhancing elusive but important clues in text discovered by the language model themselves. We found that larger language models pay more attention to non-obvious but important clues, which are often overlooked by smaller language models. Therefore, we can identify these clues by contrasting the attention weights of large and small language models. We use the identified clues as a guide to perform token-dropout data augmentation on the training text, and observed a significant boost in both small and large models' performance in fact memorization. This shows that the behavior contrast between more and less-performant language models contains important clues for knowledge learning, and it can be ``amplified"for a straight-forward improvement in knowledge learning efficiency.

Problem

• Improve knowledge learning efficiency in language models
• Address overfitting and long-distance dependency issues
• Enhance elusive clues using attention contrast between models

Innovation

• Contrast attention weights of large and small models
• Use token-dropout data augmentation on training text
• Enhance elusive clues for improved knowledge learning