Research map / Architectures and efficiency / Quantization
Optimal Brain Compression: A Framework for Accurate Post-Training Quantization and Pruning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace AdaPrune with ExactOBS (exact one-weight-at-a-time OBS) | layer-wise pruning · ResNet50 4x FLOP reduction | accuracy: 72.39 -> 74.05 | Helped |
| replace AdaPrune with ExactOBS | layer-wise pruning · BERT SQuAD 4x FLOP reduction | F1: 18.75 -> 82.10 | Helped |
| replace AdaPrune with ExactOBS | layer-wise pruning · YOLOv5l 2x FLOP reduction | mAP@0.5 | No clear effect |
| ExactOBS 2:4 instead of AdaPrune 4:8 | layer-wise pruning · ResNet50 N:M sparsity | accuracy: 74.75 -> 74.71 | No clear effect |
| use ExactOBS instead of AdaPrune with 2:4 | layer-wise pruning · SQuAD 2:4 sparsity BERT | F1: 85.24 -> 86.77 | Helped |
| OBQ instead of AdaQuant (independent layer-wise) | layer-wise quantization · ResNet18 3-bit | accuracy: 59.21 -> 68.69 | Helped |
| OBQ independent vs sequential BRECQ | quantization · ResNet18 2-bit | accuracy: 64.70 -> 64.04 | Mixed |
| apply OBQ sequentially instead of independently (ablation) | OBQ · ResNet18 2-bit | accuracy: 64.04 -> 64.93 | Helped |
| add ImageNet calibration augmentations (ablation) | OBQ · ResNet18 4-bit | accuracy: 69.59 -> 69.56 | No clear effect |
| apply global AdaPrune on top of ExactOBS rather than AdaPrune | post-processing · BERT 4x | F1: 84.10 -> 86.42 | Helped |
Closest papers by meaning
- OPQ: Compressing Deep Neural Networks with One-shot Pruning-Quantization (AAAI 2021 · similarity 0.92)
- Optimal Brain Restoration for Joint Quantization and Sparsification of LLMs (ICLR 2026 · similarity 0.91)
- Automatic Neural Network Compression by Sparsity-Quantization Joint Learning: A Constrained Optimization-Based Approach (CVPR 2020 · similarity 0.90)
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.90)
- Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings (ICLR 2024 · similarity 0.88)
- MCNC: Manifold-Constrained Reparameterization for Neural Compression (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.