Summary of Using Interleaved Ensemble Unlearning to Keep Backdoors at Bay For Finetuning Vision Transformers, by Zeyu Michael Li
Using Interleaved Ensemble Unlearning to Keep Backdoors at Bay for Finetuning Vision Transformers
by Zeyu Michael Li
First submitted to arxiv on: 1 Oct 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
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 introduces Interleaved Ensemble Unlearning (IEU), a novel method for defending Vision Transformers (ViTs) against backdoor attacks. Backdoors trigger undesirable behaviors in models during inference, compromising ViT performance in security-sensitive tasks. While CNN-specific defenses exist, few are tailored to ViTs. IEU finetunes a shallow ViT on backdoored data to achieve high confidence and low confidence on clean data. The shallow ViT then acts as a “gate” to block potentially poisoned data from the defended ViT, which is added to an unlearn set for asynchronous unlearning via gradient ascent. The authors demonstrate IEU’s effectiveness against 11 state-of-the-art backdoor attacks on three datasets and showcase its versatility by applying it to different model architectures. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine a way to protect special computer models called Vision Transformers from being tricked into doing the wrong thing. This can happen when someone tries to make the model do something bad, like recognize an image as a specific object even if it’s not really that object. The authors of this paper have come up with a new method called Interleaved Ensemble Unlearning (IEU) to help protect these models. IEU makes sure the model is very confident in what it knows and less confident in what it doesn’t know. This helps keep the model safe from bad data. The authors tested IEU on three different datasets and showed that it works well against many types of bad data. |
Keywords
» Artificial intelligence » Cnn » Inference » Vit
