AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

Being Friends Instead of Adversaries: Deep Networks Learn from Data Simplified by Other Networks

Simone Marullo, Matteo Tiezzi, Marco Gori, Stefano Melacci

AAAI 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Adversarial training refinements · 1 citations (OpenAlex)

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 a shared auxiliary neural network to simplify inputs, then drop it after trainingCNN-B · MNIST-rot-back-imagetest error rate: 22.81 ± 0.36 -> 22.22 ± 0.62Improved
Use a shared auxiliary neural network to simplify inputs, then drop it after trainingFC-B · IMDBtest error rate: 13.27 ± 0.19 -> 11.93 ± 0.09Improved
Use a shared auxiliary neural network to simplify inputs, then drop it after trainingFC-B · Winestest error rate: 17.07 ± 0.11 -> 17.15 ± 0.12Hurt
Use a shared auxiliary neural network to simplify inputs, then drop it after trainingCNN-B · CIFAR-10test error rate: 29.75 ± 0.37 -> 29.00 ± 0.36Improved
Use a shared auxiliary neural network to simplify inputs, then drop it after trainingResNet18 · CIFAR-10-n10test error rate: 8.92 ± 0.23 -> 8.10 ± 0.19Improved
Increase the maximum perturbation penalty eta_max (ablation)NFT with CNN-A · MNIST-back-imagetest error rateMixed
Use a shorter simplification schedule (ablation)NFT with CNN-A · MNIST-back-imagetest error rateHurt
Vary the number of auxiliary-network features (ablation)NFT with CNN-A · MNIST-back-imagetest error rateNo clear effect
Increase the auxiliary-network learning rate beta (ablation)NFT with CNN-A · MNIST-back-imagetest error rateImproved

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