Beyond Outliers: A Data-Free Layer-wise Mixed-Precision Quantization Approach Driven by Numerical and Structural Dual-Sensitivity

• URL: http://arxiv.org/abs/2603.17354v1
• Date: Wed, 18 Mar 2026 04:30:54 GMT
• Title: Beyond Outliers: A Data-Free Layer-wise Mixed-Precision Quantization Approach Driven by Numerical and Structural Dual-Sensitivity
• Abstract summary: Layer-wise mixed-precision quantization (LMPQ) enables effective compression under extreme low-bit settings.<n>We propose NSDS, a novel calibration-free LMPQ framework driven by Numerical and Structural Dual-Sensitivity.
• Score: 42.048576085381065
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Layer-wise mixed-precision quantization (LMPQ) enables effective compression under extreme low-bit settings by allocating higher precision to sensitive layers. However, existing methods typically treat all intra-layer weight modules uniformly and rely on a single numerical property when estimating sensitivity, overlooking their distinct operational roles and structural characteristics. To address this, we propose NSDS, a novel calibration-free LMPQ framework driven by Numerical and Structural Dual-Sensitivity. Specifically, it first mechanistically decomposes each layer into distinct operational roles and quantifies their sensitivity from both numerical and structural perspectives. These dual-aspect scores are then aggregated into a unified layer-wise metric through a robust aggregation scheme based on MAD-Sigmoid and Soft-OR to guide bit allocation. Extensive experiments demonstrate that NSDS consistently achieves superior performance compared to various baselines across diverse models and downstream tasks, without relying on any calibration data.

Related papers

• Convolution-Free Holistic Multivariance Decomposition Layer for Efficient Hyperspectral Image Classification Tensor Networks [0.0]
  This work introduces the Holistic Multispectral Decomposition (HMD) framework as a novel, end-to-end differentiable neural network layer.<n>By explicitly separating independent single mode variations from cooperative higher dimensional interactions via learnable, matrix valued supports, proposed HMD-0, HMD-1 and HMD-2 approximants are optimized jointly.<n>These results demonstrate that the HMD framework provides a structurally robust substitute for traditional convolution in multidimensional HS image classification.
  arXiv  Detail & Related papers  (2026-08-17T08:17:46Z)
• SliderQuant: Accurate Post-Training Quantization for LLMs [16.025581576593783]
  Given a pre-trained high-precision language model, the sequential quantization framework treats different layers equally.<n>We propose a new PTQ framework termed Sliding-layer Quantization (SliderQuant)<n>Our method outperforms existing PTQ methods for both weight-only quantization and weight-activation quantization.
  arXiv  Detail & Related papers  (2026-03-26T10:21:38Z)
• MS-ISSM: Objective Quality Assessment of Point Clouds Using Multi-scale Implicit Structural Similarity [65.85858856481131]
  unstructured and irregular nature of point clouds poses a significant challenge for objective quality assessment (PCQA)<n>We propose the Multi-scale Implicit Structural Similarity Measurement (MS-ISSM)
  arXiv  Detail & Related papers  (2026-01-03T14:58:52Z)
• SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size [5.229694155440675]
  Large language models (LLMs) face significant computational and memory challenges.<n>We introduce SDQ-LLM: Sigma-Delta Quantization for 1-bit LLMs of any size.<n>A distinctive feature of SDQ-LLM is the continuous layer of the Over-Sampling Ratio (OSR)
  arXiv  Detail & Related papers  (2025-09-27T14:49:58Z)
• A Further Comparison of TD-DMRG and ML-MCTDH for Nonadiabatic Dynamics of Exciton Dissociation [2.0480965608306305]
  A recent study reported up to 60% discrepancy in their calculations for excitonconfiguration.<n>By revisiting the benchmark P3HT:PCBM heterojunction model, we show that the observed discrepancies arise primarily from insufficient bond dimensions.<n>Our results confirm both methods converge to numerically exact solutions when bond dimensions are adequately scaled.
  arXiv  Detail & Related papers  (2025-08-30T10:53:51Z)
• AHDMIL: Asymmetric Hierarchical Distillation Multi-Instance Learning for Fast and Accurate Whole-Slide Image Classification [51.525891360380285]
  AHDMIL is an Asymmetric Hierarchical Distillation Multi-Instance Learning framework.<n>It eliminates irrelevant patches through a two-step training process.<n>It consistently outperforms previous state-of-the-art methods in both classification performance and inference speed.
  arXiv  Detail & Related papers  (2025-08-07T07:47:16Z)
• Towards Extreme Pruning of LLMs with Plug-and-Play Mixed Sparsity [32.668409666483626]
  Existing pruning methods mainly focus on designing metrics to measure the importance of network components to guide pruning.<n>We propose an efficient method based on the trace of Fisher Information Matrix (FIM) to quantitatively measure and verify the different sensitivities across layers.<n>Based on this, we propose Mixed Sparsity Pruning (MSP) which uses a pruning-oriented evolutionary algorithm (EA) to determine the optimal sparsity levels for different layers.
  arXiv  Detail & Related papers  (2025-03-14T08:05:49Z)
• Mix-QViT: Mixed-Precision Vision Transformer Quantization Driven by Layer Importance and Quantization Sensitivity [0.0]
  Mix-QViT is an explainability-driven MPQ framework that allocates bit-widths to each layer based on two criteria.<n>For post-training quantization, we introduce a clipped channel-wise quantization method.
  arXiv  Detail & Related papers  (2025-01-10T21:36:20Z)
• SliM-LLM: Salience-Driven Mixed-Precision Quantization for Large Language Models [63.118592279833656]
  Post-training quantization (PTQ) is an effective technique for compressing large language models (LLMs)<n>We propose SliM-LLM, a salience-driven mixed-precision quantization framework that allocates bit-widths at the group-wise.<n> Experiments show that SliM-LLM achieves superior performance across various LLMs at low bit-widths.
  arXiv  Detail & Related papers  (2024-05-23T16:21:48Z)
• CSQ: Growing Mixed-Precision Quantization Scheme with Bi-level Continuous Sparsification [51.81850995661478]
  Mixed-precision quantization has been widely applied on deep neural networks (DNNs) Previous attempts on bit-level regularization and pruning-based dynamic precision adjustment during training suffer from noisy gradients and unstable convergence. We propose Continuous Sparsification Quantization (CSQ), a bit-level training method to search for mixed-precision quantization schemes with improved stability.
  arXiv  Detail & Related papers  (2022-12-06T05:44:21Z)
• Component-Wise Natural Gradient Descent -- An Efficient Neural Network Optimization [0.0]
  Natural Gradient Descent (NGD) is a second-order neural network training that preconditions gradient descent with the inverse of the Fisher Information Matrix (FIM) This paper proposes a new NGD variant algorithm named Component-Wise Natural Gradient Descent (CW-NGD) CW-NGD requires fewer iterations to converge compared to the state-of-the-art first-order and second-order methods.
  arXiv  Detail & Related papers  (2022-10-11T09:03:57Z)
• Meta Learning Low Rank Covariance Factors for Energy-Based Deterministic Uncertainty [58.144520501201995]
  Bi-Lipschitz regularization of neural network layers preserve relative distances between data instances in the feature spaces of each layer. With the use of an attentive set encoder, we propose to meta learn either diagonal or diagonal plus low-rank factors to efficiently construct task specific covariance matrices. We also propose an inference procedure which utilizes scaled energy to achieve a final predictive distribution.
  arXiv  Detail & Related papers  (2021-10-12T22:04:19Z)
• Dual-constrained Deep Semi-Supervised Coupled Factorization Network with Enriched Prior [80.5637175255349]
  We propose a new enriched prior based Dual-constrained Deep Semi-Supervised Coupled Factorization Network, called DS2CF-Net. To ex-tract hidden deep features, DS2CF-Net is modeled as a deep-structure and geometrical structure-constrained neural network. Our network can obtain state-of-the-art performance for representation learning and clustering.
  arXiv  Detail & Related papers  (2020-09-08T13:10:21Z)

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.