Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models

• URL: http://arxiv.org/abs/2502.10835v1
• Date: Sat, 15 Feb 2025 15:36:42 GMT
• Title: Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models
• Authors: Zeping Yu, Yonatan Belinkov, Sophia Ananiadou,
• Abstract summary: We investigate how large language models perform latent multi-hop reasoning in prompts like "Wolfgang Amadeus Mozart's mother's spouse is"<n>We find that failures often stem from the relation attribute extraction stage, where conflicting logits reduce prediction accuracy.<n>We propose back attention, a novel mechanism that enables lower layers to leverage higher-layer hidden states from different positions during attention computation.
• Score: 51.53835083483751
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate how large language models perform latent multi-hop reasoning in prompts like "Wolfgang Amadeus Mozart's mother's spouse is". To analyze this process, we introduce logit flow, an interpretability method that traces how logits propagate across layers and positions toward the final prediction. Using logit flow, we identify four distinct stages in single-hop knowledge prediction: (A) entity subject enrichment, (B) entity attribute extraction, (C) relation subject enrichment, and (D) relation attribute extraction. Extending this analysis to multi-hop reasoning, we find that failures often stem from the relation attribute extraction stage, where conflicting logits reduce prediction accuracy. To address this, we propose back attention, a novel mechanism that enables lower layers to leverage higher-layer hidden states from different positions during attention computation. With back attention, a 1-layer transformer achieves the performance of a 2-layer transformer. Applied to four LLMs, back attention improves accuracy on five reasoning datasets, demonstrating its effectiveness in enhancing latent multi-hop reasoning ability.

Related papers

• Mitigating Visual Forgetting via Take-along Visual Conditioning for Multi-modal Long CoT Reasoning [53.790502697674754]
  We propose Take-along Visual Conditioning (TVC), a strategy that shifts image input to critical reasoning stages. TVC helps the model retain attention to the visual components throughout the reasoning. Our approach achieves state-of-the-art performance on average across five mathematical reasoning benchmarks.
  arXiv  Detail & Related papers  (2025-03-17T16:45:12Z)
• In-Context Linear Regression Demystified: Training Dynamics and Mechanistic Interpretability of Multi-Head Softmax Attention [52.159541540613915]
  We study how multi-head softmax attention models are trained to perform in-context learning on linear data. Our results reveal that in-context learning ability emerges from the trained transformer as an aggregated effect of its architecture and the underlying data distribution.
  arXiv  Detail & Related papers  (2025-03-17T02:00:49Z)
• VOILA: Evaluation of MLLMs For Perceptual Understanding and Analogical Reasoning [63.0285363282581]
  Multimodal Large Language Models (MLLMs) have become a powerful tool for integrating visual and textual information. We introduce VOILA, a benchmark designed to evaluate MLLMs' perceptual understanding and abstract relational reasoning. We reveal that current MLLMs struggle to comprehend inter-image relationships and exhibit limited capabilities in high-level relational reasoning.
  arXiv  Detail & Related papers  (2025-02-25T23:36:19Z)
• Unveiling Reasoning Thresholds in Language Models: Scaling, Fine-Tuning, and Interpretability through Attention Maps [3.8936716676293917]
  This study investigates the in-context learning capabilities of various decoder-only transformer-based language models with different model sizes and training data. We identify a critical parameter threshold (1.6 billion), beyond which reasoning performance improves significantly in tasks such as commonsense reasoning in multiple-choice question answering and deductive reasoning.
  arXiv  Detail & Related papers  (2025-02-21T00:48:32Z)
• Towards Interpreting Language Models: A Case Study in Multi-Hop Reasoning [0.0]
  Language models (LMs) struggle to perform multi-hop reasoning consistently. We propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on LM attention heads.
  arXiv  Detail & Related papers  (2024-11-06T16:30:26Z)
• Seemingly Plausible Distractors in Multi-Hop Reasoning: Are Large Language Models Attentive Readers? [6.525065859315515]
  We investigate whether Large Language Models (LLMs) are prone to exploiting simplifying cues in multi-hop reasoning benchmarks. Motivated by this finding, we propose a challenging multi-hop reasoning benchmark, by generating seemingly plausible multi-hop reasoning chains. We find that their performance to perform multi-hop reasoning is affected, as indicated by up to 45% relative decrease in F1 score when presented with such seemingly plausible alternatives.
  arXiv  Detail & Related papers  (2024-09-08T19:22:58Z)
• An Analysis and Mitigation of the Reversal Curse [70.13419502543915]
  Recent research observed a noteworthy phenomenon in large language models (LLMs) The reversal curse is that when dealing with two entities, $a$ and $b$, LLMs excel in handling sequences in the form of $aRb$,'' but encounter challenges when processing $bR-1a$''
  arXiv  Detail & Related papers  (2023-11-13T17:01:12Z)
• Towards a Mechanistic Interpretation of Multi-Step Reasoning Capabilities of Language Models [107.07851578154242]
  Language models (LMs) have strong multi-step (i.e., procedural) reasoning capabilities. It is unclear whether LMs perform tasks by cheating with answers memorized from pretraining corpus, or, via a multi-step reasoning mechanism. We show that MechanisticProbe is able to detect the information of the reasoning tree from the model's attentions for most examples.
  arXiv  Detail & Related papers  (2023-10-23T01:47:29Z)
• In-Context Convergence of Transformers [63.04956160537308]
  We study the learning dynamics of a one-layer transformer with softmax attention trained via gradient descent. For data with imbalanced features, we show that the learning dynamics take a stage-wise convergence process.
  arXiv  Detail & Related papers  (2023-10-08T17:55:33Z)
• Memory Injections: Correcting Multi-Hop Reasoning Failures during Inference in Transformer-Based Language Models [4.343604069244352]
  We propose an approach to pinpoint and rectify multi-hop reasoning failures through targeted memory injections on attention heads. We show that a simple, efficient, and targeted memory injection into a key attention layer can often increase the probability of the desired next token in multi-hop tasks, by up to 424%.
  arXiv  Detail & Related papers  (2023-09-11T16:39:30Z)
• Ladder-of-Thought: Using Knowledge as Steps to Elevate Stance Detection [73.31406286956535]
  We introduce the Ladder-of-Thought (LoT) for the stance detection task. LoT directs the small LMs to assimilate high-quality external knowledge, refining the intermediate rationales produced. Our empirical evaluations underscore LoT's efficacy, marking a 16% improvement over GPT-3.5 and a 10% enhancement compared to GPT-3.5 with CoT on stance detection task.
  arXiv  Detail & Related papers  (2023-08-31T14:31:48Z)
• Scalable Multi-Hop Relational Reasoning for Knowledge-Aware Question Answering [35.40919477319811]
  We propose a novel knowledge-aware approach that equips pre-trained language models with a multi-hop relational reasoning module. It performs multi-hop, multi-relational reasoning over subgraphs extracted from external knowledge graphs. It unifies path-based reasoning methods and graph neural networks to achieve better interpretability and scalability.
  arXiv  Detail & Related papers  (2020-05-01T23:10:26Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.