Summary of Influence-based Attributions Can Be Manipulated, by Chhavi Yadav et al.
Influence-based Attributions can be Manipulated
by Chhavi Yadav, Ruihan Wu, Kamalika Chaudhuri
First submitted to arxiv on: 8 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Medium Difficulty summary: This research paper presents a threat to the reliability of Influence Functions, a widely used tool for attributing predictions to training data. The authors demonstrate that an adversary can systematically tamper with influence-based attributions in logistic regression models trained on ResNet feature embeddings and standard tabular fairness datasets. They provide efficient attacks with backward-friendly implementations, raising concerns about the trustworthiness of influence-based attributions in adversarial scenarios. The paper highlights the importance of evaluating the robustness of Influence Functions against potential manipulation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This study shows that a powerful tool used to understand how data was used to make predictions can be tricked by someone trying to manipulate the results. The authors test this on different types of datasets and models, and show that it is possible to intentionally distort the attributions. They also provide ways to do this in an efficient way. This raises questions about whether we can really trust these attribution methods. |
Keywords
» Artificial intelligence » Logistic regression » Resnet
