Loading Now

Summary of Mitigating Privacy Risk in Membership Inference by Convex-concave Loss, By Zhenlong Liu et al.


Mitigating Privacy Risk in Membership Inference by Convex-Concave Loss

by Zhenlong Liu, Lei Feng, Huiping Zhuang, Xiaofeng Cao, Hongxin Wei

First submitted to arxiv on: 8 Feb 2024

Categories

  • Main: Machine Learning (cs.LG)
  • Secondary: Cryptography and Security (cs.CR)

     Abstract of paper      PDF of paper


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 proposes a novel approach to defend machine learning models against membership inference attacks (MIAs), where an attacker tries to determine whether a sample is part of the training dataset. The existing methods use gradient ascent to increase the loss variance, but this can lead to instability and poor performance. The authors introduce Convex-Concave Loss (CCL), which uses a concave term to reduce the convexity of loss functions during training. This leads to high-variance losses for training data, making it more difficult for attackers to infer membership. The CCL method is theoretically motivated and experimentally shown to achieve state-of-the-art balance between privacy and utility.
Low GrooveSquid.com (original content) Low Difficulty Summary
This paper helps keep machine learning models safe from sneaky attacks that try to figure out which samples were used to train the model. Right now, the best way to defend against these attacks involves making the training data more unpredictable. But this can make the model perform poorly or be unstable. The new method called Convex-Concave Loss (CCL) tries to find a balance between keeping the data unpredictable and having the model work well. It does this by adding a special term to the loss function that makes it harder for attackers to figure out which samples were used.

Keywords

* Artificial intelligence  * Inference  * Loss function  * Machine learning