Summary of Exploiting the Data Gap: Utilizing Non-ignorable Missingness to Manipulate Model Learning, by Deniz Koyuncu et al.
Exploiting the Data Gap: Utilizing Non-ignorable Missingness to Manipulate Model Learning
by Deniz Koyuncu, Alex Gittens, Bülent Yener, Moti Yung
First submitted to arxiv on: 6 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This paper investigates Adversarial Missingness (AM) attacks, which maliciously engineer non-ignorable missingness mechanisms to mislead causal structure learning algorithms. Existing AM attacks assume full-information maximum likelihood methods for handling missing data but are limited in applicability when alternative strategies are used. The authors propose a novel probabilistic approximation by deriving the asymptotic forms of complete case analysis, mean imputation, and regression-based imputation. They formulate the learning of the adversarial missingness mechanism as a bi-level optimization problem and demonstrate AM attacks can change p-values from significant to insignificant in real datasets like California-housing using moderate amounts of missingness (<20%). The paper also assesses the robustness of these attacks against defense strategies based on data valuation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research looks at ways that people can make machines learn incorrectly by hiding information. When machines try to figure out what’s going on, they might not get it right if some important details are missing. The researchers found a way to do this intentionally and showed that it can be very effective in making the machine think something is true when it’s actually false. They tested this idea with real data from California and found that it worked surprisingly well. This could have big implications for how we use machines to make decisions. |
Keywords
» Artificial intelligence » Likelihood » Optimization » Regression
