Research map / Architectures and efficiency / Quantization
Jumping through Local Minima: Quantization in the Loss Landscape of Vision Transformers
Natalia Frumkin, Dibakar Gope, Diana Marculescu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 gradient optimizers with block-wise evolutionary search (ablation) | FQ-ViT 8W8A · ImageNet ViT-B | top-1 accuracy: 83.40 -> 84.40 (SGD vs Evol-Q) | Improved |
| use infoNCE instead of MSE, cosine, KL loss | Evol-Q · ImageNet ViT-B | top-1 accuracy | Improved |
| apply Evol-Q on FQ-ViT for 3-bit weights | ViT-B · ImageNet 3W8A | top-1 accuracy: 55.33 -> 65.63 | Improved |
| apply Evol-Q on FQ-ViT for 4-bit Swin-T | Swin-T · ImageNet 4W8A | top-1 accuracy: 80.73 -> 80.43 | Hurt |
| apply Evol-Q on FQ-ViT for LeViT | LeViT-128S · ImageNet 4W8A | top-1 accuracy: 14.90 -> 29.20 | Improved |
| apply Evol-Q on top of BRECQ | BRECQ · ImageNet 4W4A ResNet-50 | top-1 accuracy: 75.05 -> 77.30 | Improved |
| increase calibration set to 2,000 or 5,000 images | Evol-Q · ImageNet calibration size | top-1 accuracy | Hurt |
| increase passes P from 1 to 10 | Evol-Q · ImageNet | top-1 accuracy | Improved |
| add Evol-Q on top of OMSE and bias correction (ablation) | FQ-ViT 3W8A · ImageNet ViT-B | top-1 accuracy: 73.27 -> 75.00 | Improved |
| run Evol-Q with PSAQ-ViT-V2 baseline comparison on DeiT-Tiny | DeiT-T 8W8A · ImageNet | top-1 accuracy: 72.17 vs 71.63 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PTQ4ViT — PTQ4ViT: Post-Training Quantization for Vision Transformers with Twin Uniform Quantization (ECCV 2022) · better on 1 of 1 shared result
- BRECQ — BRECQ: Pushing the Limit of Post-Training Quantization by Block Reconstruction (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GradQ-ViT: Robust and Efficient Gradient Quantization for Vision Transformers (AAAI 2025 · similarity 0.90)
- GPLQ: A General, Practical, and Lightning QAT Method for Vision Transformers (NeurIPS 2025 · similarity 0.90)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (CVPR 2025 · similarity 0.89)
- Bi-ViT: Pushing the Limit of Vision Transformer Quantization (AAAI 2024 · similarity 0.89)
- FIMA-Q: Post-Training Quantization for Vision Transformers by Fisher Information Matrix Approximation (CVPR 2025 · similarity 0.88)
- AIQViT: Architecture-Informed Post-Training Quantization for Vision Transformers (AAAI 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.
