Summary of Psbd: Prediction Shift Uncertainty Unlocks Backdoor Detection, by Wei Li et al.
PSBD: Prediction Shift Uncertainty Unlocks Backdoor Detection
by Wei Li, Pin-Yu Chen, Sijia Liu, Ren Wang
First submitted to arxiv on: 9 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 method, Prediction Shift Backdoor Detection (PSBD), tackles backdoor attacks in deep neural networks by identifying suspicious training data. This uncertainty-based approach requires minimal unlabeled clean validation data. The novel method is motivated by the Prediction Shift (PS) phenomenon, where poisoned models’ predictions on clean data often shift away from true labels towards certain other labels with dropout applied during inference. PSBD identifies backdoor training samples by computing the Prediction Shift Uncertainty (PSU), the variance in probability values when dropout layers are toggled on and off during model inference. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Backdoor attacks in deep neural networks can be a problem because they manipulate model predictions by adding malicious data to the training set. Scientists have been trying to find ways to identify this kind of data, but it’s still a challenge. This research proposes a new way to do it called Prediction Shift Backdoor Detection (PSBD). PSBD uses an idea called “prediction shift” where clean data is used to test if the model has been poisoned. The method is easy to use and doesn’t need much extra information. |
Keywords
» Artificial intelligence » Dropout » Inference » Probability
