AI Paper Atlas

Research map / Architectures and efficiency / Pruning and efficient tuning

A Bregman Learning Framework for Sparse Neural Networks

Leon Bungert, Tim Roith, Daniel Tenbrinck, Martin Burger

NeurIPS 2022 · Architectures and efficiency › Pruning and efficient tuning · approach: Sparse training from scratch

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
add momentum (beta=0.9) to LinBreg (ablation)LinBreg · Fashion-MNIST CNN, l1test accuracy: 89.2 -> 89.9Improved
use AdaBreg (Adam-style bias correction) instead of LinBreg (ablation)LinBreg · Fashion-MNIST CNN, l1test accuracy: 89.2 -> 90.5Improved
use AdaBreg instead of LinBreg (ablation)LinBreg · CIFAR-10 ResNet-18test accuracy: 90.9 -> 92.3Improved
apply post-training thresholding to LinBreg (ablation)LinBreg · CIFAR-10 ResNet-18test accuracy and sparsity: 90.9 -> 90.2Mixed
apply post-training thresholding to AdaBreg (ablation)AdaBreg · CIFAR-10 ResNet-18test accuracy and sparsity: 92.3 -> 90.5Mixed
add momentum (beta=0.9) to LinBreg (ablation)LinBreg · CIFAR-10 ResNet-18test accuracy: 90.9 -> 90.4No clear effect
Bregman (AdaBreg) vs iterative pruningiterative pruning · Fashion-MNIST CNN, l1N_total: 4.7 -> 2.3Improved
Bregman (AdaBreg) vs iterative pruning under group sparsitypruning · Fashion-MNIST CNN, group sparsitytest accuracy: 86.9 -> 89.4Improved
LinBreg vs ProxGD LassoProxGD Lasso · Fashion-MNIST CNN, l1N_total: 4.8 -> 1.9Improved
SGD + thresholding instead of ProxGD for LassoLasso · Fashion-MNIST CNN, l1test accuracy: 89.4 -> 90.1Improved
use sparse initialization with 1% non-zero entries (r=0.01)LinBreg · MNIST MLPsparsity trajectoryImproved
increase l1 regularization parameter lambdaLinBreg · MNIST MLPvalidation accuracy vs sparsity: 80% weights -> 10% weightsMixed

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