Loading...

Out-of-Distribution Generalization in Image Data with a Focus on Stable Relations

Hosseini Noohdani, Fahimeh | 2024

5 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56975 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Soleymani Baghshah, Mahdieh
  7. Abstract:
  8. One of the main challenges in the field of Machine Learning, which has come to attention in recent years, is the models' inability to generalize well to datapoints that come from a distribution different from the one that the model has been trained on. This problem is known by Out-of-Distribution Generalization. This problem is of utmost importance as in many real-world applications the training and test data do not come from the same distribution, and thus the model will fail on the test data despite its acceptable performance on samples from the same distribution of the training set. One of the main reasons behind such failure is models' reliance on spurious correlations between some features of the training datapoints and the label. One of the main challenges in several proposed methods for mitigating spurious correlation is that they require the knowledge of the value of the spurious features to train a robust model. In this project, we propose a method that by decomposing images into two causal and non-causal components, combines components of different images to mitigate spurious correlation in image classification tasks without requiring values of the spurious features. The proposed method has an overall higher average and worst test group accuracy on the Waterbirds, CelebA, Dominoes, and Metashift benchmarks compared to previous works with the same amount of supervision
  9. Keywords:
  10. Causality ; Out-Of-Distribution Generalization ; Compositional Generalization ; Spurious Correlation

 Digital Object List

 Bookmark

...see more