DART-ing Through the Drift: Dynamic Tracing of Knowledge Neurons for Adaptive Inference-Time Pruning

• URL: http://arxiv.org/abs/2601.22632v1
• Date: Fri, 30 Jan 2026 06:48:16 GMT
• Title: DART-ing Through the Drift: Dynamic Tracing of Knowledge Neurons for Adaptive Inference-Time Pruning
• Authors: Abhishek Tyagi, Yunuo Cen, Shrey Dhorajiya, Bharadwaj Veeravalli, Xuanyao Fong,
• Abstract summary: We introduce DART, a lightweight training-free method that performs on-the-fly context-based pruning.<n>DART monitors shifts in distributions to infer context changes, dynamically updating neuron-level masks to retain salient parameters.<n>It achieves accuracy gains of up to 14.5% on LLAMA-3.1-8B at 70% FFN sparsity and 3x better ROUGE-L scores with respect to static-masked pruning.
• Score: 6.3691159627915015
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Language Models (LLMs) exhibit substantial parameter redundancy, particularly in Feed-Forward Networks (FFNs). Existing pruning methods suffer from two primary limitations. First, reliance on dataset-specific calibration introduces significant data dependency and computational overhead. Second, being predominantly static, they fail to account for the evolving subset of knowledge neurons in LLMs during autoregressive generation as the context evolves. To address this, we introduce DART, i.e., Dynamic Attention-Guided Runtime Tracing), a lightweight, training-free method that performs on-the-fly context-based pruning. DART monitors shifts in attention score distributions to infer context changes, dynamically updating neuron-level masks to retain salient parameters. Across ten benchmarks, DART outperforms prior dynamic baseline, achieving accuracy gains of up to 14.5% on LLAMA-3.1-8B at 70% FFN sparsity. Furthermore, DART achieves up to 3x better ROUGE-L scores with respect to static-masked pruning on summarization tasks, with its performance comparable to the original dense models. We conclusively demonstrate that the proposed framework effectively adapts to diverse semantic contexts, preserves model capabilities across both general and domain-specific tasks while running at less than 10MBs of memory for LLAMA-3.1-8B(16GBs) with 0.1% FLOPs overhead. The code is available at https://github.com/seeder-research/DART.

Related papers

• CREAM: Continual Retrieval on Dynamic Streaming Corpora with Adaptive Soft Memory [19.64051996386645]
  CREAM is a self-supervised framework for memory-based continual retrieval.<n>It adapts to both seen and unseen topics in an unsupervised setting.<n> Experiments on two benchmark datasets demonstrate that CREAM exhibits superior adaptability and retrieval accuracy.
  arXiv  Detail & Related papers  (2026-01-06T04:47:49Z)
• DynaPURLS: Dynamic Refinement of Part-aware Representations for Skeleton-based Zero-Shot Action Recognition [51.80782323686666]
  We introduce textbfDynaPURLS, a unified framework that establishes robust, multi-scale visual-semantic correspondences.<n>Our framework leverages a large language model to generate hierarchical textual descriptions that encompass both global movements and local body-part dynamics.<n>Experiments on three large-scale benchmark datasets, including NTU RGB+D 60/120 and PKU-MMD, demonstrate that DynaPURLS significantly outperforms prior art.
  arXiv  Detail & Related papers  (2025-12-12T10:39:10Z)
• DART: Differentiable Dynamic Adaptive Region Tokenizer for Vision Foundation Models [45.12546316524245]
  We introduce DART, a fully differentiable Dynamic Region Adaptive Tokenizer.<n>DART employs learnable region scores and quantile-based partitioning to create content-aware patches of varying sizes.<n>The impact of this approach is profound, where a DART-Small matches the performance of a DeiT-Base86 with nearly double the inference speed.
  arXiv  Detail & Related papers  (2025-06-12T06:25:37Z)
• SuperFlow++: Enhanced Spatiotemporal Consistency for Cross-Modal Data Pretraining [62.433137130087445]
  SuperFlow++ is a novel framework that integrates pretraining and downstream tasks using consecutive camera pairs.<n>We show that SuperFlow++ outperforms state-of-the-art methods across diverse tasks and driving conditions.<n>With strong generalizability and computational efficiency, SuperFlow++ establishes a new benchmark for data-efficient LiDAR-based perception in autonomous driving.
  arXiv  Detail & Related papers  (2025-03-25T17:59:57Z)
• PRISM: Self-Pruning Intrinsic Selection Method for Training-Free Multimodal Data Selection [68.8373788348678]
  Visual instruction tuning adapts pre-trained Multimodal Large Language Models to follow human instructions.<n>PRISM is the first training-free framework for efficient visual instruction selection.<n>It reduces the end-to-end time for data selection and model tuning to just 30% of conventional pipelines.
  arXiv  Detail & Related papers  (2025-02-17T18:43:41Z)
• Dual-frame Fluid Motion Estimation with Test-time Optimization and Zero-divergence Loss [9.287932323337163]
  3D particle tracking velocimetry (PTV) is a key technique for analyzing turbulent flow. Deep learning-based methods have achieved impressive accuracy in dual-frame fluid motion estimation. We introduce a new method that is completely self-supervised and notably outperforms its fully-supervised counterparts.
  arXiv  Detail & Related papers  (2024-10-15T18:00:00Z)
• Dynamic Graph Message Passing Networks for Visual Recognition [112.49513303433606]
  Modelling long-range dependencies is critical for scene understanding tasks in computer vision. A fully-connected graph is beneficial for such modelling, but its computational overhead is prohibitive. We propose a dynamic graph message passing network, that significantly reduces the computational complexity.
  arXiv  Detail & Related papers  (2022-09-20T14:41:37Z)
• SD-Conv: Towards the Parameter-Efficiency of Dynamic Convolution [16.56592303409295]
  Dynamic convolution achieves better performance for efficient CNNs at the cost of negligible FLOPs increase. We propose a new framework, textbfSparse Dynamic Convolution (textscSD-Conv), to naturally integrate these two paths.
  arXiv  Detail & Related papers  (2022-04-05T14:03:54Z)
• DS-Net++: Dynamic Weight Slicing for Efficient Inference in CNNs and Transformers [105.74546828182834]
  We show a hardware-efficient dynamic inference regime, named dynamic weight slicing, which adaptively slice a part of network parameters for inputs with diverse difficulty levels. We present dynamic slimmable network (DS-Net) and dynamic slice-able network (DS-Net++) by input-dependently adjusting filter numbers of CNNs and multiple dimensions in both CNNs and transformers.
  arXiv  Detail & Related papers  (2021-09-21T09:57:21Z)
• LiDAR-based Panoptic Segmentation via Dynamic Shifting Network [56.71765153629892]
  LiDAR-based panoptic segmentation aims to parse both objects and scenes in a unified manner. We propose the Dynamic Shifting Network (DS-Net), which serves as an effective panoptic segmentation framework in the point cloud realm. Our proposed DS-Net achieves superior accuracies over current state-of-the-art methods.
  arXiv  Detail & Related papers  (2020-11-24T08:44:46Z)

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.