mlsanityData to 7 Oct 2026

Solving Oscillation Problem in Post-Training Quantization Through a Theoretical Perspective

arXiv 2303.11906 ↗First posted 21 Mar 2023cs.CVcited by 17

Not on the map. Why?

Abstract (arXiv)

Post-training quantization (PTQ) is widely regarded as one of the most efficient compression methods practically, benefitting from its data privacy and low computation costs. We argue that an overlooked problem of oscillation is in the PTQ methods. In this paper, we take the initiative to explore and present a theoretical proof to explain why such a problem is essential in PTQ. And then, we try to solve this problem by introducing a principled and generalized framework theoretically. In particular, we first formulate the oscillation in PTQ and prove the problem is caused by the difference in module capacity. To this end, we define the module capacity (ModCap) under data-dependent and data-free scenarios, where the differentials between adjacent modules are used to measure the degree of oscillation. The problem is then solved by selecting top-k differentials, in which the corresponding modules are jointly optimized and quantized. Extensive experiments demonstrate that our method successfully reduces the performance drop and is generalized to different neural networks and PTQ methods. For example, with 2/4 bit ResNet-50 quantization, our method surpasses the previous state-of-the-art method by 1.9%. It becomes more significant on small model quantization, e.g. surpasses BRECQ method by 6.61% on MobileNetV2*0.5.

Key facts. Cited by 17 papers in the map (data through 7 Oct 2026). Builds on 37 of its 47 references that are in the map. First posted on arXiv 21 Mar 2023.

Builds on

37 of 47 references.

References · 37

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  3. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,441
  4. 2015Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shiftcited by 6,979
  5. 2018MobileNetV2: Inverted Residuals and Linear Bottleneckscited by 4,773
  6. 2015Deep Compression: Compressing Deep Neural Networks with Pruning, Trained Quantization and Huffman Codingcited by 2,291
  7. 2018DARTS: Differentiable Architecture Searchcited by 1,782
  8. 2016Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learningcited by 1,509
  9. 2020Designing Network Design Spacescited by 954
  10. 2019Learned Step Size Quantizationcited by 604
  11. 2020Up or Down? Adaptive Rounding for Post-Training Quantizationcited by 600
  12. 2018Quantizing deep convolutional networks for efficient inference: A whitepapercited by 507
25 more
  1. 2017Progressive Neural Architecture Searchcited by 483
  2. 2021BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstructioncited by 464
  3. 2018PACT: Parameterized Clipping Activation for Quantized Neural Networkscited by 456
  4. 2018HAQ: Hardware-Aware Automated Quantization with Mixed Precisioncited by 426
  5. 2020ZeroQ: A Novel Zero Shot Quantization Frameworkcited by 265
  6. 2019Differentiable Soft Quantization: Bridging Full-Precision and Low-Bit Neural Networkscited by 234
  7. 2018Bi-Real Net: Enhancing the Performance of 1-bit CNNs With Improved Representational Capability and Advanced Training Algorithmcited by 233
  8. 2022QDrop: Randomly Dropping Quantization for Extremely Low-bit Post-Training Quantizationcited by 175
  9. 2019Loss Aware Post-training Quantizationcited by 94
  10. 2020Improving Post Training Neural Quantization: Layer-wise Calibration and Integer Programmingcited by 89
  11. 2019Towards Unified INT8 Training for Convolutional Neural Networkcited by 77
  12. 2017Towards Effective Low-bitwidth Convolutional Neural Networkscited by 76
  13. 2022Revisiting Random Channel Pruning for Neural Network Compressioncited by 50
  14. 2021When to Prune? A Policy towards Early Structural Pruningcited by 38
  15. 2021OMPQ: Orthogonal Mixed Precision Quantizationcited by 37
  16. 2021Fire Together Wire Together: A Dynamic Pruning Approach with Self-Supervised Mask Predictioncited by 24
  17. 2019Multinomial Distribution Learning for Effective Neural Architecture Searchcited by 17
  18. 2021An Information Theory-inspired Strategy for Automatic Network Pruningcited by 17
  19. 2022Interspace Pruning: Using Adaptive Filter Representations to Improve Training of Sparse CNNscited by 16
  20. 2021You Only Compress Once: Towards Effective and Elastic BERT Compression via Exploit-Explore Stochastic Nature Gradientcited by 14
  21. 2019DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruningcited by 13
  22. 2021Do All MobileNets Quantize Poorly? Gaining Insights into the Effect of Quantization on Depthwise Separable Convolutional Networks Through the Eyes of Multi-scale Distributional Dynamicscited by 12
  23. 2020Progressive Automatic Design of Search Space for One-Shot Neural Architecture Searchcited by 10
  24. 2020Rethinking Performance Estimation in Neural Architecture Searchcited by 7
  25. 2017Take it in your stride: Do we need striding in CNNs?cited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 1; 2024: 3; 2025: 3; 2026: 10.

17 papers.

Sort:

By area:

  1. 2024AffineQuant: Affine Transformation Quantization for Large Language Modelscited by 64
  2. 2024PTQ4SAM: Post-Training Quantization for Segment Anythingcited by 32
  3. 2025Low-bit Model Quantization for Deep Neural Networks: A Surveycited by 11
  4. 2023EPTQ: Enhanced Post-Training Quantization via Hessian-guided Network-wise Optimizationcited by 5
  5. 2024MetaAug: Meta-Data Augmentation for Post-Training Quantizationcited by 5
  6. 2026Motion-Aware Caching for Efficient Autoregressive Video Generationcited by 2
  7. 2025Pack-PTQ: Advancing Post-training Quantization of Neural Networks by Pack-wise Reconstructioncited by 1
  8. 2025Efficiently Training A Flat Neural Network Before It has been Quantizatedcited by 1
  9. 2026CAR-SAM: Cross-Attention Reconstruction for Post-Training Quantization of the Segment Anything Modelcited by 1
  10. 2026ALGOGEN: Tool-Generated Verifiable Traces for Reliable Algorithm Visualizationcited by 0
  11. 2026A2RBench: An Automatic Paradigm for Formally Verifiable Abstract Reasoning Benchmark Generationcited by 0
  12. 2026InTrain: Intrinsic Trainability for Zero-Cost Neural Architecture Searchcited by 0
5 more
  1. 2026FUSE: Frequency-domain Unification and Spectral Energy Alignment for Multi-modal Object Re-Identificationcited by 0
  2. 2026Efficient Tuning Before Low-Bit Post-Training Quantization for Stochastic Gradient Descent-optimized Modelscited by 0
  3. 2026CommitKV: Lifecycle-Aware KV Cache Compression via Commit Transitions for Multi-Turn Agentscited by 0
  4. 2026UC-VLM: Consistency-Driven Learning for AI-Generated Image Detection with Vision-Language Large Modelscited by 0
  5. 2026STEPQuant: When and Where Errors Matter in Delta-Rule Recurrent State Quantizationcited by 0