networks with prior knowledge
play

networks with prior knowledge Laura Rieger Chandan Singh W. James - PowerPoint PPT Presentation

Interpretations are useful: penalizing explanations to align neural networks with prior knowledge Laura Rieger Chandan Singh W. James Murdoch Bin Yu DTU UC Berkeley UC Berkeley UC Berkeley overview datasets are biased Benign NNs


  1. Interpretations are useful: penalizing explanations to align neural networks with prior knowledge Laura Rieger Chandan Singh W. James Murdoch Bin Yu DTU UC Berkeley UC Berkeley UC Berkeley

  2. overview

  3. datasets are biased Benign • NNs learn from large datasets • often biased • we sometimes know the bias Cancerous

  4. augmenting the loss function Prediction True label Explanation Prior knowledge

  5. using our method improves accuracy Image Vanilla Our method more focus on skin less focus on band-aid Test F1: 0.67 0.73

  6. details

  7. Learning from labels (step by step) training with biased data Benign  90% accurate Cancerous

  8. what did the network learn? Benign  Cancerous

  9. We know the bias (sometimes) Gender is not important for job applications! Race shouldn’t determine jail time! Rulers aren’t cancerous! Band aids don’t protect against cancer!

  10. our method

  11. augmenting the loss function Prediction True label

  12. augmenting the loss function Prediction True label Explanation Prior knowledge

  13. C ontextual D ecomposition E xplanation P enalty any differentiable explanation method works we used contextual decomposition (Singh 2019) captures interactions computationally lighter [1] Singh, Chandan, W. James Murdoch, and Bin Yu. "Hierarchical interpretations for neural network predictions." 13

  14. Contextual Decomposition (Singh 2019) • requires partition of input • iteratively forward-pass both partitions • output contribution of both partitions

  15. results

  16. skin cancer (ISIC) explanations focus more on skin

  17. mnist variants

  18. contributions

  19. contributions CDEP uses explainability methods to regularize an NN used to incorporate prior knowledge into neural networks 0.67 (f1) 0.73 (f1) usable with more complex unpenalized penalized knowledge than previous methods

Recommend


More recommend