mlsanityData to 14 Oct 2026

On sensitivity of meta-learning to support data

arXiv 2110.13953 ↗First posted 26 Oct 2021cs.LGcited by 15

Not on the map. Why?

Abstract (arXiv)

Meta-learning algorithms are widely used for few-shot learning. For example, image recognition systems that readily adapt to unseen classes after seeing only a few labeled examples. Despite their success, we show that modern meta-learning algorithms are extremely sensitive to the data used for adaptation, i.e. support data. In particular, we demonstrate the existence of (unaltered, in-distribution, natural) images that, when used for adaptation, yield accuracy as low as 4% or as high as 95% on standard few-shot image classification benchmarks. We explain our empirical findings in terms of class margins, which in turn suggests that robust and safe meta-learning requires larger margins than supervised learning.

Key facts. Cited by 15 papers in the map (data through 14 Oct 2026). Builds on 32 of its 49 references that are in the map. First posted on arXiv 26 Oct 2021.

Builds on

32 of 49 references.

References · 32

  1. 2014Explaining and Harnessing Adversarial Examplescited by 7,364
  2. 2017Towards Deep Learning Models Resistant to Adversarial Attackscited by 6,415
  3. 2017Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networkscited by 4,897
  4. 2017Prototypical Networks for Few-shot Learningcited by 3,015
  5. 2016Towards Evaluating the Robustness of Neural Networkscited by 2,998
  6. 2016Matching Networks for One Shot Learningcited by 2,255
  7. 2020FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidencecited by 2,085
  8. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,375
  9. 2017Targeted Backdoor Attacks on Deep Learning Systems Using Data Poisoningcited by 1,055
  10. 2018On First-Order Meta-Learning Algorithmscited by 1,018
  11. 2020Meta-Learning in Neural Networks: A Surveycited by 950
  12. 2017OptNet: Differentiable Optimization as a Layer in Neural Networkscited by 599
20 more
  1. 2019Meta-Learning with Differentiable Convex Optimizationcited by 461
  2. 2018TADAM: Task dependent adaptive metric for improved few-shot learningcited by 408
  3. 2018Meta-learning with differentiable closed-form solverscited by 397
  4. 2019A Baseline for Few-Shot Image Classificationcited by 298
  5. 2017Few-Shot Image Recognition by Predicting Parameters from Activationscited by 122
  6. 2018Learning to Learn How to Learn: Self-Adaptive Visual Navigation Using Meta-Learningcited by 93
  7. 2019Generalized Inner Loop Meta-Learningcited by 93
  8. 2019Meta-Curvaturecited by 70
  9. 2020learn2learn: A Library for Meta-Learning Researchcited by 69
  10. 2019Investigating Meta-Learning Algorithms for Low-Resource Natural Language Understanding Taskscited by 48
  11. 2020Unraveling Meta-Learning: Understanding Feature Representations for Few-Shot Taskscited by 47
  12. 2020Rapidly Adaptable Legged Robots via Evolutionary Meta-Learningcited by 43
  13. 2020Data Augmentation for Meta-Learningcited by 38
  14. 2019Meta Learning for End-to-End Low-Resource Speech Recognitioncited by 29
  15. 2019Fairness Warnings and Fair-MAML: Learning Fairly with Minimal Datacited by 23
  16. 2018Adversarial Meta-Learningcited by 16
  17. 2020Is Support Set Diversity Necessary for Meta-Learning?cited by 14
  18. 2018Training Medical Image Analysis Systems like Radiologistscited by 13
  19. 2020Yet Meta Learning Can Adapt Fast, It Can Also Break Easilycited by 10
  20. 2017Meta learning Framework for Automated Drivingcited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 5; 2023: 4; 2024: 2; 2025: 3; 2026: 1.

15 papers.

Sort:

By area:

  1. 2025PolyPythias: Stability and Outliers across Fifty Language Model Pre-Training Runscited by 35
  2. 2022Channel Importance Matters in Few-Shot Image Classificationcited by 20
  3. 2022Alleviating the Sample Selection Bias in Few-shot Learning by Removing Projection to the Centroidcited by 20
  4. 2023A Survey on Stability of Learning with Limited Labelled Data and its Sensitivity to the Effects of Randomnesscited by 14
  5. 2023From Channel Bias to Feature Redundancy: Uncovering the "Less is More" Principle in Few-Shot Learningcited by 10
  6. 2022Learning to Detect Noisy Labels Using Model-Based Featurescited by 6
  7. 2022Log-Euclidean Signatures for Intrinsic Distances Between Unaligned Datasetscited by 5
  8. 2024Automatic Combination of Sample Selection Strategies for Few-Shot Learningcited by 5
  9. 2025Deciphering Trajectory-Aided LLM Reasoning: An Optimization Perspectivecited by 2
  10. 2023Exploring Active Learning in Meta-Learning: Enhancing Context Set Labelingcited by 1
  11. 2024On Sensitivity of Learning with Limited Labelled Data to the Effects of Randomness: Impact of Interactions and Systematic Choicescited by 1
  12. 2026Benchmarking Few-shot Transferability of Pre-trained Models with Improved Evaluation Protocolscited by 1
3 more
  1. 2022Class-Specific Channel Attention for Few-Shot Learningcited by 0
  2. 2023Evaluating the Evaluators: Are Current Few-Shot Learning Benchmarks Fit for Purpose?cited by 0
  3. 2025TLXML: Task-Level Explanation of Meta-Learning via Influence Functionscited by 0