Authors: Jinghan Yang, Linjie Xu, Lequan Yu

Published on: May 22, 2023

Impact Score: 8.3

Arxiv code: Arxiv:2305.12809

Summary

• What is new: First research investigating the minimum training subset needed to flip a machine learning model’s prediction.
• Why this is important: Understanding why a machine learning model predicts unsatisfactorily and how to reverse this outcome.
• What the research proposes: An efficient algorithm using an extended influence function for binary classification models to identify and relabel the necessary training subset.
• Results: Relabeling fewer than 2% of training points can flip a prediction, showing a method for challenging model predictions and assessing model robustness.

Technical Details

Technological frameworks used: Extended influence function

Models used: Binary classification models with convex loss

Data used: nan

Potential Impact

AI-driven companies, especially those involved in machine learning model development and deployment, finance, healthcare, and security sectors.

Want to implement this idea in a business?

We have generated a startup concept here: FlipPoint AI.