Summary of Demystifying Sgd with Doubly Stochastic Gradients, by Kyurae Kim et al.
Demystifying SGD with Doubly Stochastic Gradients
by Kyurae Kim, Joohwan Ko, Yi-An Ma, Jacob R. Gardner
First submitted to arxiv on: 3 Jun 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Machine Learning (cs.LG); Optimization and Control (math.OC)
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 paper investigates the convergence properties of doubly stochastic gradient descent (doubly SGD), a popular optimization strategy for summing intractable expectations. Doubly SGD is used in diffusion models, variational autoencoders, and other applications where infinite data is involved. Despite its popularity, little was known about doubly SGD’s convergence under general conditions. This work establishes the convergence of doubly SGD with independent minibatching and random reshuffling (RR) under general conditions, including dependent component gradient estimators. The analysis suggests that RR improves the complexity dependence on subsampling noise. By understanding where to invest a per-iteration computational budget, this paper provides insights for optimizing doubly SGD in various applications. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about making sure a special kind of computer program (called doubly stochastic gradient descent) works correctly and efficiently. This program is important because it helps with big tasks like creating new images or understanding language. The authors wanted to know if this program would always get the right answer, even when there’s a lot of information involved. They figured out that it will work correctly most of the time, as long as they use certain techniques to make sure the calculations are accurate. This is important because it helps people create better programs and improve how computers do tasks. |
Keywords
» Artificial intelligence » Optimization » Stochastic gradient descent
