Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed method for label shift estimation incrementally updates the prior on each sample to adjust each posterior for more accurate estimation.method 90%
• The proposed method relies on a weaker notion of calibration compared to other label shift estimation methods.technical 85%
• The proposed method is versatile and can be applied to any black-box probabilistic classifier as a post-hoc approach.technical 90%
• Experiments on CIFAR-10 and MNIST show the proposed method consistently outperforms current state-of-the-art maximum likelihood-based methods.result 85%
• The method performs well under different calibrations and varying intensities of label shift.result 80%
• The proposed method is based on intuitive assumptions on classifiers that are generally true for modern probabilistic classifiers.technical 80%
• The method differs from previous approaches that perform moment matching with confusion matrix estimated from a validation set or maximize likelihood with expectation-maximization algorithms.method 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card