Emergence of Primacy and Recency Effect in Mamba: A Mechanistic Point of View

• URL: http://arxiv.org/abs/2506.15156v1
• Date: Wed, 18 Jun 2025 06:02:02 GMT
• Title: Emergence of Primacy and Recency Effect in Mamba: A Mechanistic Point of View
• Authors: Muhammad Cendekia Airlangga, Hilal AlQuabeh, Munachiso S Nwadike, Kentaro Inui,
• Abstract summary: We study memory in state-space language models using primacy and recency effects as behavioral tools to uncover how information is retained and forgotten over time.<n>Applying structured recall tasks to the Mamba architecture, we observe a consistent U-shaped accuracy profile, indicating strong performance at the beginning and end of input sequences.
• Score: 16.8179962093575
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We study memory in state-space language models using primacy and recency effects as behavioral tools to uncover how information is retained and forgotten over time. Applying structured recall tasks to the Mamba architecture, we observe a consistent U-shaped accuracy profile, indicating strong performance at the beginning and end of input sequences. We identify three mechanisms that give rise to this pattern. First, long-term memory is supported by a sparse subset of channels within the model's selective state space block, which persistently encode early input tokens and are causally linked to primacy effects. Second, short-term memory is governed by delta-modulated recurrence: recent inputs receive more weight due to exponential decay, but this recency advantage collapses when distractor items are introduced, revealing a clear limit to memory depth. Third, we find that memory allocation is dynamically modulated by semantic regularity: repeated relations in the input sequence shift the delta gating behavior, increasing the tendency to forget intermediate items. We validate these findings via targeted ablations and input perturbations on two large-scale Mamba-based language models: one with 1.4B and another with 7B parameters.

Related papers

• From Verbatim to Gist: Distilling Pyramidal Multimodal Memory via Semantic Information Bottleneck for Long-Horizon Video Agents [78.30630000529133]
  We propose MM-Mem, a pyramidal multimodal memory architecture grounded in Fuzzy-Trace Theory.<n> MM-Mem memory structures hierarchically into a Sensory Buffer, Episodic Stream, and Symbolic.<n>Experiments confirm the effectiveness of MM-Mem on both offline and streaming tasks.
  arXiv  Detail & Related papers  (2026-03-02T05:12:45Z)
• Memory Caching: RNNs with Growing Memory [56.25483647131372]
  We introduce Memory Caching (MC), a technique that enhances recurrent models by caching checkpoints of memory states (a.k.a. hidden states)<n>We propose four variants of MC, including gated aggregation and sparse selective mechanisms, and discuss their implications on both linear and deep memory modules.<n>The results indicate that while Transformers achieve the best accuracy, our MC variants show competitive performance, close the gap with Transformers, and performs better than state-of-the-art recurrent models.
  arXiv  Detail & Related papers  (2026-02-27T18:53:41Z)
• FlashMem: Distilling Intrinsic Latent Memory via Computation Reuse [4.210760734549566]
  FlashMem is a framework that distills intrinsic memory directly from transient reasoning states via computation reuse.<n>Experiments demonstrate that FlashMem matches the performance of heavy baselines while reducing inference latency by 5 times.
  arXiv  Detail & Related papers  (2026-01-09T03:27:43Z)
• MemMamba: Rethinking Memory Patterns in State Space Model [6.537535831000493]
  We show that selective state-space models such as Mamba have high efficiency with O(n) time and O(1) recurrent inference, yet their long-range memory decays exponentially.<n>Inspired by how humans distill and retain salient information when reading long documents, we propose MemMamba.<n>MemMamba achieves significant improvements over existing Mamba variants and Transformers on long-sequence benchmarks.
  arXiv  Detail & Related papers  (2025-09-28T14:40:58Z)
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free [81.65559031466452]
  We conduct experiments to investigate gating-augmented softmax attention variants.<n>We find that a simple modification-applying a head-specific sigmoid gate after the Scaled Dot-Product Attention (SDPA)-consistently improves performance.
  arXiv  Detail & Related papers  (2025-05-10T17:15:49Z)
• MsMemoryGAN: A Multi-scale Memory GAN for Palm-vein Adversarial Purification [40.80205521005344]
  We propose a novel defense model named MsMemoryGAN to filter the perturbations from adversarial samples before recognition. MsMemoryGAN learns to reconstruct the input by merely using fewer prototypical elements of the normal patterns recorded in the memory. Our approach removes a wide variety of adversarial perturbations, allowing vein classifiers to achieve the highest recognition accuracy.
  arXiv  Detail & Related papers  (2024-08-20T09:46:30Z)
• CItruS: Chunked Instruction-aware State Eviction for Long Sequence Modeling [52.404072802235234]
  We introduce Chunked Instruction-aware State Eviction (CItruS), a novel modeling technique that integrates the attention preferences useful for a downstream task into the eviction process of hidden states. Our training-free method exhibits superior performance on long sequence comprehension and retrieval tasks over several strong baselines under the same memory budget.
  arXiv  Detail & Related papers  (2024-06-17T18:34:58Z)
• Adversarially Diversified Rehearsal Memory (ADRM): Mitigating Memory Overfitting Challenge in Continual Learning [0.0]
  Continual learning focuses on learning non-stationary data distribution without forgetting previous knowledge. Rehearsal-based approaches are commonly used to combat catastrophic forgetting. We introduce the Adversarially Diversified Rehearsal Memory to address the memory overfitting challenge.
  arXiv  Detail & Related papers  (2024-05-20T06:56:43Z)
• Simple linear attention language models balance the recall-throughput tradeoff [60.06020449520365]
  We propose BASED, a simple architecture combining linear and sliding window attention.<n>We train language models up to 1.3b parameters and show that BASED matches the strongest sub-quadratic models in perplexity and outperforms them on real-world recall-intensive tasks by 6.22 accuracy points.
  arXiv  Detail & Related papers  (2024-02-28T19:28:27Z)
• Recurrent Action Transformer with Memory [39.58317527488534]
  This paper proposes a novel model architecture that incorporates a recurrent memory mechanism designed to regulate information retention. We conduct experiments on memory-intensive environments (ViZDoom-Two-Colors, T-Maze, Memory Maze, Minigrid-Memory), classic Atari games, and MuJoCo control environments. The results show that using memory can significantly improve performance in memory-intensive environments, while maintaining or improving results in classic environments.
  arXiv  Detail & Related papers  (2023-06-15T19:29:08Z)
• Multi-level Memory-augmented Appearance-Motion Correspondence Framework for Video Anomaly Detection [1.9511777443446219]
  We propose a multi-level memory-augmented appearance-motion correspondence framework. The latent correspondence between appearance and motion is explored via appearance-motion semantics alignment and semantics replacement training. Our framework outperforms the state-of-the-art methods, achieving AUCs of 99.6%, 93.8%, and 76.3% on UCSD Ped2, CUHK Avenue, and ShanghaiTech datasets.
  arXiv  Detail & Related papers  (2023-03-09T08:43:06Z)
• LaMemo: Language Modeling with Look-Ahead Memory [50.6248714811912]
  We propose Look-Ahead Memory (LaMemo) that enhances the recurrence memory by incrementally attending to the right-side tokens. LaMemo embraces bi-directional attention and segment recurrence with an additional overhead only linearly proportional to the memory length. Experiments on widely used language modeling benchmarks demonstrate its superiority over the baselines equipped with different types of memory.
  arXiv  Detail & Related papers  (2022-04-15T06:11:25Z)
• Adaptive Online Incremental Learning for Evolving Data Streams [4.3386084277869505]
  The first major difficulty is concept drift, that is, the probability distribution in the streaming data would change as the data arrives. The second major difficulty is catastrophic forgetting, that is, forgetting what we have learned before when learning new knowledge. Our research builds on this observation and attempts to overcome these difficulties.
  arXiv  Detail & Related papers  (2022-01-05T14:25:53Z)
• HM4: Hidden Markov Model with Memory Management for Visual Place Recognition [54.051025148533554]
  We develop a Hidden Markov Model approach for visual place recognition in autonomous driving. Our algorithm, dubbed HM$4$, exploits temporal look-ahead to transfer promising candidate images between passive storage and active memory. We show that this allows constant time and space inference for a fixed coverage area.
  arXiv  Detail & Related papers  (2020-11-01T08:49:24Z)
• Memformer: A Memory-Augmented Transformer for Sequence Modeling [55.780849185884996]
  We present Memformer, an efficient neural network for sequence modeling. Our model achieves linear time complexity and constant memory space complexity when processing long sequences.
  arXiv  Detail & Related papers  (2020-10-14T09:03:36Z)

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.