Summary of Hidden Activations Are Not Enough: a General Approach to Neural Network Predictions, by Samuel Leblanc et al.
Hidden Activations Are Not Enough: A General Approach to Neural Network Predictions
by Samuel Leblanc, Aiky Rasolomanana, Marco Armenta
First submitted to arxiv on: 20 Sep 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Neural and Evolutionary Computing (cs.NE); Representation Theory (math.RT)
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 We present a novel mathematical framework for analyzing neural networks using quiver representation theory. This approach enables quantification of the similarity between a new data sample and training data as perceived by the network. By leveraging induced quiver representations, we capture more information than traditional hidden layer outputs. Our results are architecture-agnostic and task-agnostic, making them broadly applicable to various MLP architectures and adversarial attack methods. We demonstrate our framework’s effectiveness on MNIST and FashionMNIST datasets for detecting adversarial examples using publicly available code. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine a new way to understand how neural networks work. Our team created a special mathematical tool that helps us see what a network thinks about a new piece of data, compared to the old training data it learned from. This tool is super powerful because it doesn’t care what kind of network or problem we’re trying to solve. We tested this idea on two famous datasets and showed that it can help detect when someone is trying to trick the network with fake data. |
