Data Attribution Insights
Andrew discusses the intriguing relationship between influence functions and regularization when applied to randomly projected vectors. He highlights the innovative fTrace dataset from MIT, which serves as a benchmark for evaluating data attribution methods in language models. Despite improvements over existing methods, the findings reveal that traditional information retrieval systems still outperform data attribution techniques, prompting further exploration into the underlying reasons.In this clip
From this podcast
Machine Learning Street Talk (MLST)
Adversarial Examples and Data Modelling - Andrew Ilyas (MIT)
Related Questions