Targeted Recovery of Weight-Space Mechanisms From Neural Networks

• URL: http://arxiv.org/abs/2607.13047v1
• Date: Fri, 19 Jun 2026 19:07:38 GMT
• Title: Targeted Recovery of Weight-Space Mechanisms From Neural Networks
• Abstract summary: decomposition (PD) decomposes neural networks into interpretable computational components.<n>We propose targeted PD (tPD), which identifies only the components that process specific inputs of interest.<n>We validate tPD on toy models and on transformer language models trained on The Pile.
• Score: 3.4333309114070105
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Parameter decomposition (PD) decomposes neural networks into interpretable computational components that faithfully reflect the original network's operations. However, scaling PD to large models requires vast compute, making it a costly and risky endeavor. Here we propose targeted PD (tPD), which identifies only the components that process specific inputs of interest -- from isolated prompts to large subtasks -- by introducing a high-rank catch-all component that handles all non-target data. We validate tPD on toy models and on transformer language models trained on The Pile, where it recovers reproducible, mechanistically faithful circuits. We extract a CSS-only submodel of a 4-block transformer using 7% of the FLOPs of its published decomposition, and in a 12-block transformer we surgically ablate and rewire memorized sequences, with negligible side effects on other inputs.

Related papers

• Multi-layer MIMO Relay as Deep Physical Neural Networks: Power Amplifiers as Activation Functions [56.870614369768354]
  Wireless physical neural networks (WPNNs) embed neural computation directly into analog hardware.<n>We propose a deep WPNN in which nonlinear activations are realized by a multi-hop multiple-input multiple-output (MIMO) relay network.<n>We show that the proposed architecture enables accurate over-the-air inference for image classification.
  arXiv  Detail & Related papers  (2026-07-20T09:45:07Z)
• Self-Purification Mitigates Backdoors in Multimodal Diffusion Language Models [74.1970982768771]
  We show that well-established data-poisoning pipelines can successfully implant backdoors into MDLMs.<n>We introduce a backdoor defense framework for MDLMs named DiSP (Diffusion Self-Purification)
  arXiv  Detail & Related papers  (2026-02-24T15:47:52Z)
• REFINE: Inversion-Free Backdoor Defense via Model Reprogramming [60.554146386198376]
  Backdoor attacks on deep neural networks (DNNs) have emerged as a significant security threat.<n>We propose REFINE, an inversion-free backdoor defense method based on model reprogramming.
  arXiv  Detail & Related papers  (2025-02-22T07:29:12Z)
• Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition [0.0]
  We introduce a conceptual foundation for Attribution-based Decomposition (APD)<n>APD directly decomposes a neural network's parameters into components that are faithful to the parameters of the original network.<n>We demonstrate APD's effectiveness by successfully identifying ground truth mechanisms in toy experimental settings.
  arXiv  Detail & Related papers  (2025-01-24T21:31:12Z)
• CFPFormer: Feature-pyramid like Transformer Decoder for Segmentation and Detection [1.837431956557716]
  We propose Cross Feature Pyramid Transformer decoder (CFPFormer), a novel decoder block that integrates feature pyramids and transformers.<n>Our work is capable of capturing long-range dependencies and effectively up-sample feature maps.<n>With a ResNet50 backbone, our method achieves 92.02% Dice Score, highlighting the efficacy of our methods.
  arXiv  Detail & Related papers  (2024-04-23T18:46:07Z)
• Mapping of attention mechanisms to a generalized Potts model [50.91742043564049]
  We show that training a neural network is exactly equivalent to solving the inverse Potts problem by the so-called pseudo-likelihood method. We also compute the generalization error of self-attention in a model scenario analytically using the replica method.
  arXiv  Detail & Related papers  (2023-04-14T16:32:56Z)
• Integral Migrating Pre-trained Transformer Encoder-decoders for Visual Object Detection [78.2325219839805]
  imTED improves the state-of-the-art of few-shot object detection by up to 7.6% AP. Experiments on MS COCO dataset demonstrate that imTED consistently outperforms its counterparts by 2.8%.
  arXiv  Detail & Related papers  (2022-05-19T15:11:20Z)
• SATr: Slice Attention with Transformer for Universal Lesion Detection [39.90420943500884]
  Universal Lesion Detection (ULD) in computed tomography plays an essential role in computer-aided diagnosis. We propose a novel Slice Attention Transformer (SATr) block which can be easily plugged into convolution-based ULD backbones. Experiments with five state-of-the-art methods show that the proposed SATr block can provide an almost free boost to lesion detection accuracy.
  arXiv  Detail & Related papers  (2022-03-13T03:37:27Z)
• DBIA: Data-free Backdoor Injection Attack against Transformer Networks [6.969019759456717]
  We propose DBIA, a data-free backdoor attack against the CV-oriented transformer networks. Our approach can embed backdoors with a high success rate and a low impact on the performance of the victim transformers.
  arXiv  Detail & Related papers  (2021-11-22T08:13:51Z)
• PnP-DETR: Towards Efficient Visual Analysis with Transformers [146.55679348493587]
  Recently, DETR pioneered the solution vision tasks with transformers, it directly translates the image feature map into the object result. Recent transformer-based image recognition model andTT show consistent efficiency gain.
  arXiv  Detail & Related papers  (2021-09-15T01:10:30Z)
• Transformers Solve the Limited Receptive Field for Monocular Depth Prediction [82.90445525977904]
  We propose TransDepth, an architecture which benefits from both convolutional neural networks and transformers. This is the first paper which applies transformers into pixel-wise prediction problems involving continuous labels.
  arXiv  Detail & Related papers  (2021-03-22T18:00:13Z)

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.