Interaction-Aware Influence Functions for Group Attribution
Learn to apply interaction-aware influence functions to better understand how groups of training examples affect model performance, which is crucial for improving model reliability and fairness
- Build a dataset with grouped training examples
- Apply the standard influence function to estimate individual influences
- Configure the interaction-aware influence function to capture joint effects
- Test the new function on a held-out set to evaluate its effectiveness
- Apply the interaction-aware influence function to real-world datasets to analyze group attributions
Data scientists and machine learning engineers can benefit from this technique to analyze and improve their models, especially when working with complex datasets and group attributions
💡 The standard practice of summing individual influences does not capture joint effects, but interaction-aware influence functions can distinguish between redundant and complementary examples
💡 Improve model reliability with interaction-aware influence functions!
Key Takeaways
Learn to apply interaction-aware influence functions to better understand how groups of training examples affect model performance, which is crucial for improving model reliability and fairness
DeepCamp AI