Summary of Redex: Beyond Fixed Representation Methods Via Convex Optimization, by Amit Daniely et al.
RedEx: Beyond Fixed Representation Methods via Convex Optimization
by Amit Daniely, Mariano Schain, Gilad Yehudai
First submitted to arxiv on: 15 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Optimization and Control (math.OC); 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 paper aims to bridge the gap between optimizing neural networks and achieving provable optimization guarantees for fixed representation methods like kernels and random features. The proposed architecture, RedEx (Reduced Expander Extractor), combines the expressiveness of neural networks with the trainability of fixed representation methods via a convex program with semi-definite constraints. This allows for layer-wise training with optimization guarantees. Moreover, RedEx is shown to surpass fixed representation methods in terms of learning target functions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper tries to make two things work well together: making neural networks better and making sure we can prove that they’re working correctly. Neural networks are great at doing certain tasks, but they’re hard to understand and improve. On the other hand, simpler methods like kernels and random features do have guarantees, but they don’t perform as well. The new idea called RedEx tries to mix these two things together by being as good as neural networks and also being trainable in a way that we can prove is working correctly. |
Keywords
* Artificial intelligence * Optimization
