Summary of How Free Is Parameter-free Stochastic Optimization?, by Amit Attia et al.
How Free is Parameter-Free Stochastic Optimization?
by Amit Attia, Tomer Koren
First submitted to arxiv on: 5 Feb 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 The paper investigates whether there exist optimization methods that can achieve competitive convergence rates without requiring prior knowledge of problem parameters. Existing approaches are only partially parameter-free, as they rely on some assumptions about the true problem. The authors demonstrate a simple hyperparameter search technique that outperforms state-of-the-art algorithms in the non-convex setting and provide a similar result for the convex setting with noisy function values. However, they also establish a lower bound showing that fully parameter-free stochastic convex optimization is infeasible, and propose a partially parameter-free method that approaches this limit. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper explores whether there’s an optimization method that can work well without knowing much about the problem beforehand. Current methods are only “partly” good at this, as they assume some things about the true problem. The researchers found a simple way to search for good settings that outperforms more complex methods in some cases. They also showed that it’s not possible to have an optimization method that works perfectly without any prior knowledge, but they did come up with a method that gets close. |
Keywords
* Artificial intelligence * Hyperparameter * Optimization
