Summary of Timber! Poisoning Decision Trees, by Stefano Calzavara et al.
Timber! Poisoning Decision Trees
by Stefano Calzavara, Lorenzo Cazzaro, Massimo Vettori
First submitted to arxiv on: 1 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR); Machine Learning (stat.ML)
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 research introduces Timber, a novel white-box poisoning attack specifically designed for decision trees. The approach employs a greedy strategy and sub-tree retraining to efficiently measure the damage caused by poisoning a single training instance. A tree annotation procedure enables the sorting of instances based on computational cost, allowing for an early stopping criterion that enhances attack efficiency and feasibility on larger datasets. Additionally, the authors extend Timber to traditional random forest models, highlighting the value of combining decision trees into ensembles. Experimental evaluations on public datasets demonstrate superior performance compared to existing baselines in terms of effectiveness, efficiency, or both. While two representative defenses can mitigate the effect of Timber, they fail to effectively thwart the attacks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This research creates a new way to harm decision tree models by intentionally adding bad information during training. They call this attack “Timber” and it’s different from other attacks because it works directly with the model’s internal workings. The approach is efficient and can be used on large datasets. The researchers also show how to extend Timber to work with random forest models, which are often used together. Tests on public data sets show that Timber is better than previous methods at causing harm or being efficient. Some defense strategies can slow down Timber, but they’re not effective. |
Keywords
* Artificial intelligence * Decision tree * Early stopping * Random forest
