AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Jumping through Local Minima: Quantization in the Loss Landscape of Vision Transformers

Natalia Frumkin, Dibakar Gope, Diana Marculescu

ICCV 2023 · Architectures and efficiency › Quantization · approach: Post-training quantization of ViTs · 18 citations (OpenAlex)

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 changedWhereMeasuredEffect
replace gradient optimizers with block-wise evolutionary search (ablation)FQ-ViT 8W8A · ImageNet ViT-Btop-1 accuracy: 83.40 -> 84.40 (SGD vs Evol-Q)Improved
use infoNCE instead of MSE, cosine, KL lossEvol-Q · ImageNet ViT-Btop-1 accuracyImproved
apply Evol-Q on FQ-ViT for 3-bit weightsViT-B · ImageNet 3W8Atop-1 accuracy: 55.33 -> 65.63Improved
apply Evol-Q on FQ-ViT for 4-bit Swin-TSwin-T · ImageNet 4W8Atop-1 accuracy: 80.73 -> 80.43Hurt
apply Evol-Q on FQ-ViT for LeViTLeViT-128S · ImageNet 4W8Atop-1 accuracy: 14.90 -> 29.20Improved
apply Evol-Q on top of BRECQBRECQ · ImageNet 4W4A ResNet-50top-1 accuracy: 75.05 -> 77.30Improved
increase calibration set to 2,000 or 5,000 imagesEvol-Q · ImageNet calibration sizetop-1 accuracyHurt
increase passes P from 1 to 10Evol-Q · ImageNettop-1 accuracyImproved
add Evol-Q on top of OMSE and bias correction (ablation)FQ-ViT 3W8A · ImageNet ViT-Btop-1 accuracy: 73.27 -> 75.00Improved
run Evol-Q with PSAQ-ViT-V2 baseline comparison on DeiT-TinyDeiT-T 8W8A · ImageNettop-1 accuracy: 72.17 vs 71.63Hurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy