Summary of Batch Size Invariant Adam, by Xi Wang et al.
Batch size invariant Adam
by Xi Wang, Laurence Aitchison
First submitted to arxiv on: 29 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 proposes a new version of Adam, a popular optimization algorithm used in large-scale distributed settings. The innovation is a batch size invariant variant that divides mini-batches into micro-batches and distributes them among worker nodes. This allows for better performance and scalability in scenarios where previous approaches have limitations. The proposed scheme does not require strong assumptions about gradient variance, making it more practical and applicable to a broader range of use cases. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper introduces a new way to make Adam work well in big computer networks. Normally, when we train models on lots of data, we split the data into smaller groups and give each group to many computers to process at the same time. The problem is that some old methods don’t work as well with this approach. This paper shows how to fix one of these old methods by changing how it averages the information from all those small groups. This new method works better than the old one in a lot more situations. |
Keywords
* Artificial intelligence * Optimization
