Work

Statistical and Computational Guarantees for Influence Diagnostics

Year: 2022

Type: preprint

Abstract: Influence diagnostics such as influence functions and approximate maximum influence perturbations are popular in machine learning and in AI domain applications. Influence diagnostics are powerful statistical tools to identify influential datapoints or subsets of datapoints. We establish finite-sample statistical bounds, as well as computational complexity bounds, for influence functions and approximate maximum influence perturbations using efficient inverse-Hessian-vector product implementations. We illustrate our results with generalized linear models and large attention based models on synthetic and real data.

Source: arXiv (Cornell University)

DOI: 10.48550/arxiv.2212.04014

Authors: Fisher, Jillian, Liu, Lang, Pillutla, Krishna, Choi, Yejin, Harchaoui, Zaid

Institutions: University of Washington, Allen Institute for Artificial Intelligence

Language: English

Topic: Advanced Statistical Methods and Models

Subfield: Statistics and Probability

Field: Mathematics

Domain: Physical Sciences

Open Access status: green

Funders: National Science Foundation, Canadian Institute for Advanced Research, University of Washington, National Institutes of Health, Simons Institute for the Theory of Computing, University of California Berkeley

Awards: Pacific Interdisciplinary Hub on Optimal Transport, Entropic Regularization of Optimal Transport, G4243306591, TRIPODS: Institute for Foundations of Data Science, G683845418, AI Institute: Institute for Foundations of Machine Learning, Scaling Laws of Deep Learning

OpenAlex ID: W4310997702

Cites:

Cited by:

Related to: 10