Infusion: Shaping Model Behavior by Editing Training Data via Influence Functions

1. 1Identify the desired model behavior change
2. 2Compute influence functions to attribute model behavior to training documents
3. 3Apply scalable influence-function approximations to find small perturbations to training documents
4. 4Evaluate the effectiveness of Infusion on data poisoning tasks across different domains

ML researchers and engineers on a team can benefit from Infusion to refine model behavior, while data scientists can apply it to improve model performance

💡 Influence functions can be used in reverse to craft training data that induces targeted model behavior

🚀 Infusion: edit training data to shape model behavior with influence functions!