AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

A Robust Optimization Guided Pruning Framework for Vision and Large Language Models

Gabriel Afriat, Hussein Hazimeh, Dimitris Paparas, Rahul Mazumder

ICML 2026 · Architectures and efficiency › Pruning and efficient tuning · approach: Structured network pruning

Read the paper

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
use Frobenius-norm uncertainty set (adds gamma*I) instead of default tiny dampeningCAP on DeiT-Base · ImageNet-1K, 80% sparsityaccuracy: 45.19 -> 55.19Improved
use Frobenius-norm uncertainty setCAP on DeiT-Small · ImageNet-1K, 70% sparsityaccuracy: 53.77 -> 60.04Improved
use Frobenius-norm uncertainty setCAP on DeiT-Tiny · ImageNet-1K, 80% sparsityaccuracy: 8.22 -> 8.34No clear effect
use Frobenius-norm uncertainty setCAP on DeiT-Base · ImageNet-1K, 2:4 sparsityaccuracy: 75.29 -> 77.34Improved
use eigenvalue-wise uncertainty setALPS on Llama-3.1-8B · WikiText2, 70% sparsityperplexity: 30.16 -> 29.53No clear effect
use Frobenius-norm uncertainty setALPS on Llama-3.1-8B · WikiText2, 80% sparsityperplexity: 113.18 -> 105.01Improved
use Frobenius-norm uncertainty setALPS on Llama-2-13b-hf · PTB, 80% sparsityperplexity: 503.36 -> 529.50Hurt
use eigenvalue-wise uncertainty setALPS on Llama-3.1-8B · pruning time, 60% sparsityminutes: 53.97 -> 56.52Hurt
use eigenvalue-wise uncertainty setCAP on DeiT-Base · pruning timeminutes: 45.58 -> 81.51Hurt
use Frobenius-norm uncertainty set (gamma-based)ALPS on Llama-3.1-8B · mean accuracy of 3 tasks, 70% sparsityaccuracy: 48.86 -> 48.99No clear effect
use gradient-proportional uncertainty set instead of constant-radius (ablation)RobOP-CAP on DeiT-Base · ImageNet-1K, 80% sparsityaccuracy: 55.19 -> 47.36Hurt
tune gamma instead of default CAP dampening 1e-8 (ablation)CAP on DeiT-Base · ImageNet-1K, 80% sparsityaccuracyImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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