Summary of Persistent Backdoor Attacks in Continual Learning, by Zhen Guo et al.
Persistent Backdoor Attacks in Continual Learning
by Zhen Guo, Abhinav Kumar, Reza Tourani
First submitted to arxiv on: 20 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR)
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 The proposed paper investigates the persistence of backdoor attacks in neural networks during continual learning. The authors introduce two novel attacks: Blind Task Backdoor and Latent Task Backdoor. These attacks leverage minimal adversarial influence to manipulate model outputs on specific inputs. The study evaluates the efficacy of these attacks under various configurations, demonstrating high success rates across different continual learning algorithms. Furthermore, the paper shows that both attacks effectively evade state-of-the-art defenses. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Backdoors in neural networks can be a serious problem because they allow bad actors to control how the network behaves on specific inputs. The research focuses on how these backdoors persist when the model is updated with new data over time. Two new types of attacks are proposed: one that changes how the loss is calculated and another that only affects training for one task. The study shows that both attacks work well against different algorithms and can evade some defenses. |
Keywords
» Artificial intelligence » Continual learning
