Summary of Ensemble Everything Everywhere: Multi-scale Aggregation For Adversarial Robustness, by Stanislav Fort et al.
Ensemble everything everywhere: Multi-scale aggregation for adversarial robustness
by Stanislav Fort, Balaji Lakshminarayanan
First submitted to arxiv on: 8 Aug 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 The proposed approach to achieving high-quality representations for deep neural networks involves the use of multi-resolution input representations and dynamic self-ensembling of intermediate layer predictions. This allows for adversarial robustness without any additional training or data, with an accuracy of approximately 72% on CIFAR-10 and 48% on CIFAR-100 on the RobustBench AutoAttack suite. The method also achieves a comparable result to top models on CIFAR-10 and improves the state-of-the-art by 5% and 9% respectively on CIFAR-100. Further experiments show that simple gradient-based attacks against the model produce human-interpretable images of target classes, providing insights into the interplay between adversarial robustness and hierarchical representations. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper proposes a new way to make deep neural networks more robust to fake or “adversarial” examples. This is done by using multiple types of input data and combining predictions from different layers in the network. The approach works well on two popular datasets, CIFAR-10 and CIFAR-100, without needing any special training or extra data. By adding some simple additional training, the results get even better. |
