Locating and Editing Factual Associations in Mamba

• URL: http://arxiv.org/abs/2404.03646v2
• Date: Fri, 2 Aug 2024 21:29:25 GMT
• Title: Locating and Editing Factual Associations in Mamba
• Authors: Arnab Sen Sharma, David Atkinson, David Bau,
• Abstract summary: We investigate the mechanisms of factual recall in the Mamba state space model. We compare Mamba directly to a similar-sized autoregressive transformer LM.
• Score: 22.097117651225595
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We investigate the mechanisms of factual recall in the Mamba state space model. Our work is inspired by previous findings in autoregressive transformer language models suggesting that their knowledge recall is localized to particular modules at specific token locations; we therefore ask whether factual recall in Mamba can be similarly localized. To investigate this, we conduct four lines of experiments on Mamba. First, we apply causal tracing or interchange interventions to localize key components inside Mamba that are responsible for recalling facts, revealing that specific components within middle layers show strong causal effects at the last token of the subject, while the causal effect of intervening on later layers is most pronounced at the last token of the prompt, matching previous findings on autoregressive transformers. Second, we show that rank-one model editing methods can successfully insert facts at specific locations, again resembling findings on transformer LMs. Third, we examine the linearity of Mamba's representations of factual relations. Finally we adapt attention-knockout techniques to Mamba in order to dissect information flow during factual recall. We compare Mamba directly to a similar-sized autoregressive transformer LM and conclude that despite significant differences in architectural approach, when it comes to factual recall, the two architectures share many similarities.

Related papers

• Differential Mamba [16.613266337054267]
  Sequence models like Transformers and RNNs often overallocate attention to irrelevant context, leading to noisy intermediate representations.<n>Recent work has shown that differential design can mitigate this issue in Transformers, improving their effectiveness across various applications.<n>We show that a naive adaptation of differential design to Mamba is insufficient and requires careful architectural modifications.
  arXiv  Detail & Related papers  (2025-07-08T17:30:14Z)
• Mamba Knockout for Unraveling Factual Information Flow [22.520634805939093]
  We rely on theoretical and empirical connections to Transformer-based architectures and their attention mechanisms.<n>We adapt attentional interpretability techniques originally developed for Transformers to both Mamba-1 and Mamba-2.<n>By further leveraging Mamba's structured factorization, we disentangle how distinct "features" either enable token-to-token information exchange or enrich individual tokens.
  arXiv  Detail & Related papers  (2025-05-30T06:08:36Z)
• TransMamba: Fast Universal Architecture Adaption from Transformers to Mamba [88.31117598044725]
  We explore cross-architecture training to transfer the ready knowledge in existing Transformer models to alternative architecture Mamba, termed TransMamba. Our approach employs a two-stage strategy to expedite training new Mamba models, ensuring effectiveness in across uni-modal and cross-modal tasks. For cross-modal learning, we propose a cross-Mamba module that integrates language awareness into Mamba's visual features, enhancing the cross-modal interaction capabilities of Mamba architecture.
  arXiv  Detail & Related papers  (2025-02-21T01:22:01Z)
• From Markov to Laplace: How Mamba In-Context Learns Markov Chains [36.22373318908893]
  We study in-context learning on Markov chains and uncover a surprising phenomenon. Unlike transformers, even a single-layer Mamba efficiently learns the in-context Laplacian smoothing estimator. These theoretical insights align strongly with empirical results and represent the first formal connection between Mamba and optimal statistical estimators.
  arXiv  Detail & Related papers  (2025-02-14T14:13:55Z)
• Mimetic Initialization Helps State Space Models Learn to Recall [81.43140985343358]
  Recent work has shown that state space models such as Mamba are significantly worse than Transformers on recall-based tasks. We investigate whether their poor copying and recall performance could be due in part to training difficulties rather than fundamental capacity constraints.
  arXiv  Detail & Related papers  (2024-10-14T23:17:46Z)
• Towards Universality: Studying Mechanistic Similarity Across Language Model Architectures [49.24097977047392]
  We investigate two mainstream architectures for language modeling, namely Transformers and Mambas, to explore the extent of their mechanistic similarity. We propose to use Sparse Autoencoders (SAEs) to isolate interpretable features from these models and show that most features are similar in these two models.
  arXiv  Detail & Related papers  (2024-10-09T08:28:53Z)
• A Survey of Mamba [27.939712558507516]
  Recently, a novel architecture named Mamba has emerged as a promising alternative for building foundation models. This study investigates the advancements of Mamba-based models, the techniques of adapting Mamba to diverse data, and the applications where Mamba can excel.
  arXiv  Detail & Related papers  (2024-08-02T09:18:41Z)
• Mamba meets crack segmentation [0.18416014644193066]
  Cracks pose safety risks to infrastructure and cannot be overlooked. CNNs exhibit a deficiency in global modeling capability, hindering the representation to entire crack features. This study explores the representation capabilities of Mamba to crack features.
  arXiv  Detail & Related papers  (2024-07-22T15:21:35Z)
• An Empirical Study of Mamba-based Pedestrian Attribute Recognition [15.752464463535178]
  This paper designs and adapts Mamba into two typical PAR frameworks, text-image fusion approach and pure vision Mamba multi-label recognition framework. It is found that interacting with attribute tags as additional input does not always lead to an improvement, specifically, Vim can be enhanced, but VMamba cannot. These experimental results indicate that simply enhancing Mamba with a Transformer does not always lead to performance improvements but yields better results under certain settings.
  arXiv  Detail & Related papers  (2024-07-15T00:48:06Z)
• Venturing into Uncharted Waters: The Navigation Compass from Transformer to Mamba [77.21394300708172]
  Transformer, a deep neural network architecture, has long dominated the field of natural language processing and beyond. The recent introduction of Mamba challenges its supremacy, sparks considerable interest among researchers, and gives rise to a series of Mamba-based models that have exhibited notable potential. This survey paper orchestrates a comprehensive discussion, diving into essential research dimensions, covering: (i) the functioning of the Mamba mechanism and its foundation on the principles of structured state space models; (ii) the proposed improvements and the integration of Mamba with various networks, exploring its potential as a substitute for Transformers; (iii) the combination of
  arXiv  Detail & Related papers  (2024-06-24T15:27:21Z)
• Demystify Mamba in Vision: A Linear Attention Perspective [72.93213667713493]
  Mamba is an effective state space model with linear computation complexity. We show that Mamba shares surprising similarities with linear attention Transformer. We propose a Mamba-Like Linear Attention (MLLA) model by incorporating the merits of these two key designs into linear attention.
  arXiv  Detail & Related papers  (2024-05-26T15:31:09Z)
• ReMamber: Referring Image Segmentation with Mamba Twister [51.291487576255435]
  ReMamber is a novel RIS architecture that integrates the power of Mamba with a multi-modal Mamba Twister block. The Mamba Twister explicitly models image-text interaction, and fuses textual and visual features through its unique channel and spatial twisting mechanism.
  arXiv  Detail & Related papers  (2024-03-26T16:27:37Z)
• Is Mamba Capable of In-Context Learning? [63.682741783013306]
  State of the art foundation models such as GPT-4 perform surprisingly well at in-context learning (ICL) This work provides empirical evidence that Mamba, a newly proposed state space model, has similar ICL capabilities.
  arXiv  Detail & Related papers  (2024-02-05T16:39:12Z)

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.