AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

Task Discovery: Finding the Tasks that Neural Networks Generalize on

Andrei Atanov, Andrey Filatov, Teresa Yeo, Ajay Sohmshetty, Amir Zamir

NeurIPS 2022 · Label-efficient and robust learning › Test-time adaptation · approach: Extrapolation and single-sample adaptation · 4 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
Use an adversarial train-test split based on a randomly initialized network task instead of a random splitCIFAR-10 multi-class classification · CIFAR-10accuracy: 0.78±0.04 -> 0.41±0.10Hurt
Use an adversarial train-test split based on a randomly initialized network task instead of a random splitCelebA hair_blond classification · CelebA hair_blondaccuracy: 0.94±0.00 -> 0.42±0.02Hurt
Use an adversarial train-test split based on a randomly initialized network task instead of a random splitImageNet classification · ImageNettop-1 accuracy: 0.59±0.01 -> 0.29±0.00Hurt
Use the discovered task set instead of randomly initialized network tasksResNet-18 task discovery · CIFAR-10average task similarity: 0.68 -> 0.51Improved
Use a SimCLR-pretrained embedding as the task-discovery input (ablation)Task discovery · CIFAR-10similarity to human-labelled tasksImproved
Vary the similarity-loss weight λ (ablation)Shared-embedding task discovery · CIFAR-10average agreement scoreNo clear effect
Change any one stochasticity source between training runs (ablation)Agreement-score measurement · CIFAR-10 human- and random-labelled tasksagreement score: 1Hurt
Use fewer training images (ablation)Agreement-score measurement on human-labelled tasks · CIFAR-10agreement scoreHurt
Use fewer training images (ablation)Agreement-score measurement on random-labelled tasks · CIFAR-10agreement scoreMixed
Use the discovered tasks rather than human-labelled tasksAgreement-score measurement across test domains · CIFAR-10 training; multiple test domainsagreement score stabilityImproved
Discover tasks on TinyImageNet instead of evaluating human-labelled binary tasksResNet-18 · TinyImageNetagreement score: below 0.65 -> above 0.8Improved

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