AI Paper Atlas

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

Smoothed Embeddings for Certified Few-Shot Learning

Mikhail Pautov, Olesya Kuznetsova, Nurislam Tursynbek, Aleksandr Petiushko, Ivan Oseledets

NeurIPS 2022 · Adversarial robustness and security › Certified and empirical robustness · approach: Randomized smoothing certification · 0 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
Increase number of noise samples n from 10^3 to 10^5 (ablation)certification Algorithm · Cub-200-2011time per image (s): 0.044 -> 4.744Hurt
Increase n from 10^3 to 10^4 (ablation)certification Algorithm · Cub-200-2011time per image (s): 0.044 -> 0.509Hurt
Decrease alpha from 1e-2 to 1e-4 (ablation)smoothed ProtoNet · CIFAR-FS 5-shotnon-certified objects (%): 3.0 -> 3.8Hurt
Decrease alpha from 1e-2 to 1e-4 (ablation)smoothed ProtoNet · Cub-200-2011 1-shotnon-certified objects (%): 1.6 -> 1.6No clear effect
Decrease alpha from 1e-2 to 1e-4 (ablation)smoothed ProtoNet · miniImageNet 1-shotnon-certified objects (%): 1.9 -> 2.4Hurt
Decrease alpha from 1e-2 to 1e-4 (ablation)smoothed ProtoNet · miniImageNet 5-shotnon-certified objects (%): 2.9 -> 3.0No clear effect
Use a smaller sample size n for certification (ablation)smoothed ProtoNet · CIFAR-FS 5-shotcertified accuracyNo clear effect
Train the baseline encoder with Gaussian noise augmentation (ablation)smoothed ProtoNet · Cub-200-2011, CIFAR-FS, miniImageNetempirical robust accuracyImproved
Random additive noise attack on plain ProtoNetplain baseline · three datasetsempirical robust accuracyNo clear effect
FGSM adversarial attack on smoothed modelsmoothed ProtoNet · three datasetsempirical robust accuracyNo clear effect
Replace one-step FGSM with 20-step PGD attack (ablation)smoothed ProtoNet · three datasets 1-shotempirical robust accuracyHurt

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