Summary of Edge Of Stochastic Stability: Revisiting the Edge Of Stability For Sgd, by Arseniy Andreyev and Pierfrancesco Beneventano
Edge of Stochastic Stability: Revisiting the Edge of Stability for SGD
by Arseniy Andreyev, Pierfrancesco Beneventano
First submitted to arxiv on: 29 Dec 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 In this paper, researchers explore the connection between neural network training methods and their convergence properties. Specifically, they study how full-batch gradient descent with a certain step size affects the largest eigenvalue of the Hessian matrix. The results have important implications for understanding how models generalize well or poorly. The authors also investigate mini-batch stochastic gradient descent (SGD) and show that it operates in a distinct regime where different factors come into play. They introduce the concept of “Batch Sharpness” to describe this phenomenon, which has significant consequences for our understanding of SGD’s behavior. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how neural networks are trained. It shows that when we use a special method called full-batch gradient descent, some important numbers in the math stay stable at certain values. This is useful because it helps us understand why models work well or poorly. The researchers also look at another way to train models called mini-batch stochastic gradient descent (SGD). They find that this works differently and has its own special rules. They give a name, “Batch Sharpness”, to describe what happens when we use SGD. |
Keywords
» Artificial intelligence » Gradient descent » Neural network » Stochastic gradient descent
