Shortcut learning in deep neural networks
Explore this paper's citation graph
Summary
It is proposed that deep learning failures are a consequence of shortcut learning, a common characteristic across biological and artificial systems in which strategies that appear to have solved a problem fail unexpectedly under different circumstances.
- Type
- article
- Published
- 2020-04-16
- Cited by
- 3,389
- References
- 167
- Access
- Open access
- OpenAlex
- https://openalex.org/W3016970897
- Semantic Scholar
- https://api.semanticscholar.org/CorpusID:215786368
Keywords
Artificial intelligence, Computer science, Deep learning, Transferability, Benchmarking
References
- Predictability, surprise, attention, and conditioning
- Vision: A Computational Investigation into the Human Representation and Processing of Visual Information
- The Winograd Schema Challenge
- Intriguing properties of neural networks
- Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification
- Learning and Behavior: A Contemporary Synthesis
- An introduction to comparative psychology
- Causal inference by using invariant prediction: identification and confidence intervals
- Deep neural networks are easily fooled: High confidence predictions for unrecognizable images
- Rethinking eliminative connectionism.
- ON QUALITATIVE DIFFERENCES IN LEARNING—II OUTCOME AS A FUNCTION OF THE LEARNER'S CONCEPTION OF THE TASK
- Learning in Science: A Comparison of Deep and Surface Approaches.
- Contemporary Animal Learning Theory
- Scene context influences without scene gist: Eye movements guided by spatial associations in visual search
- Unbiased look at dataset bias
- Laser-optical investigation of stator-rotor interaction in a transonic turbine
- Animal detection in natural scenes: critical features revisited.
- Scene perception: detecting and judging objects undergoing relational violations.
- Individual differences in study processes and the Quality of Learning Outcomes
- Fairness through awareness
Cited by
- Human-like generalization in a machine through predicate learning
- Analysing Deep Reinforcement Learning Agents Trained with Domain Randomisation
- Deconfounded Image Captioning: A Causal Retrospect
- Overinterpretation reveals image classification model pathologies
- What Do Models Learn from Question Answering Datasets?
- Diverse deep neural networks all predict human IT well, after training and fitting
- On the Value of Out-of-Distribution Testing: An Example of Goodhart's Law
- Adversarial NLI for Factual Correctness in Text Summarisation Models
- Is Independence all you need? On the Generalization of Representations Learned from Correlated Data
- Roses are Red, Violets are Blue… But Should VQA expect Them To?
- Mitigating Gender Bias in Captioning Systems
- What shapes feature representations? Exploring datasets, architectures, and training
- Overcoming Statistical Shortcuts for Open-ended Visual Counting
- Time for a Background Check! Uncovering the impact of Background Features on Deep Neural Networks
- The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization
- Beyond accuracy: quantifying trial-by-trial behaviour of CNNs and humans by measuring error consistency
- Improving robustness against common corruptions by covariate shift adaptation
- Usefulness of interpretability methods to explain deep learning based plant stress phenotyping
- Challenges and Opportunities of End-to-End Learning in Medical Image Classification
- How benign is benign overfitting?
Related papers
- Defense-guided Transferable Adversarial Attacks
- Remix: Towards the transferability of adversarial examples
- Transferable Normalization: Towards Improving Transferability of Deep Neural Networks
- Boosting the Transferability of Adversarial Attacks in Deep Neural Networks
- Comprehensive Analysis of Hyperdimensional Computing Against Gradient Based Attacks
- Progressive Diversified Augmentation for General Robustness of DNNs: A Unified Approach
- A Causal View on Robustness of Neural Networks
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks
- Exploring Architectural Ingredients of Adversarially Robust Deep Neural Networks