Corpus ID: 227255149

Understanding Failures of Deep Networks via Robust Feature Extraction

@article{Singla2020UnderstandingFO,
  title={Understanding Failures of Deep Networks via Robust Feature Extraction},
  author={Sahil Singla and Besmira Nushi and S. Shah and Ece Kamar and E. Horvitz},
  journal={ArXiv},
  year={2020},
  volume={abs/2012.01750}
}
Traditional evaluation metrics for learned models that report aggregate scores over a test set are insufficient for surfacing important and informative patterns of failure over features and instances. We introduce and study a method aimed at characterizing and explaining failures by identifying visual attributes whose presence or absence results in poor performance. In distinction to previous work that relies upon crowdsourced labels for visual attributes, we leverage the representation of a… Expand

References

SHOWING 1-10 OF 46 REFERENCES
Adversarial Robustness as a Prior for Learned Representations
  • 57
  • Highly Influential
Robustness May Be at Odds with Accuracy
  • 634
  • Highly Influential
  • PDF
Towards Accountable AI: Hybrid Human-Machine Analyses for Characterizing System Failure
  • 44
  • PDF
Counterfactual Explanations for Machine Learning: A Review
  • 19
  • PDF
Feature Purification: How Adversarial Training Performs Robust Deep Learning
  • 22
  • PDF
Generative causal explanations of black-box classifiers
  • 5
  • PDF
Understanding the role of individual units in a deep neural network
  • 26
  • PDF
Visualizing the Impact of Feature Attribution Baselines
  • 30
Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
  • 418
  • PDF
Counterfactual Visual Explanations
  • 97
  • PDF
...
1
2
3
4
5
...