Summary of Certified Peftsmoothing: Parameter-efficient Fine-tuning with Randomized Smoothing, by Chengyan Fu and Wenjie Wang
Certified PEFTSmoothing: Parameter-Efficient Fine-Tuning with Randomized Smoothing
by Chengyan Fu, Wenjie Wang
First submitted to arxiv on: 8 Apr 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 Randomized smoothing is a prominent certified robustness method for assessing deep learning models’ resistance to l2-norm adversarial perturbations by adding Gaussian noise and aggregating base classifier votes. Theoretically, it provides a certified norm bound ensuring stable predictions within this bound. However, the need to retrain base models from scratch hinders widespread adoption due to the failure to learn the noise-augmented data distribution. To overcome this limitation, we propose Parameter-Efficient Fine-Tuning (PEFT) methods for adapting base models to learn Gaussian noise-augmented data in white-box and black-box settings. Our approach, PEFTSmoothing, efficiently certifies robustness with extensive results demonstrating over 98% accuracy on CIFAR-10, outperforming state-of-the-art denoised smoothing by 20%, and over 61% accuracy on ImageNet, comparable to diffusion-based methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Randomized smoothing is a way to make deep learning models more stable against changes in images. It’s like adding noise to the image and then asking multiple models what they think the correct answer is. The problem with this method is that it’s hard to train these models from scratch, which makes it difficult to use. To fix this, we came up with a new way to fine-tune existing models so they can work well with noisy images. This new approach, called PEFTSmoothing, works in two ways: one where the model knows what the noise looks like and another where it doesn’t. We tested our method on several image datasets and found that it worked really well, especially compared to other state-of-the-art methods. |
Keywords
» Artificial intelligence » Deep learning » Diffusion » Fine tuning » Parameter efficient
