Summary of Learning High-degree Parities: the Crucial Role Of the Initialization, by Emmanuel Abbe et al.
Learning High-Degree Parities: The Crucial Role of the Initialization
by Emmanuel Abbe, Elisabetta Cornacchia, Jan Hązła, Donald Kougang-Yombi
First submitted to arxiv on: 6 Dec 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 investigates the learnability of degree k parities by regular neural networks trained with gradient descent. It shows that the efficiency of learning depends on the initial weight distribution, specifically whether it is initialized with a discrete Rademacher distribution or a Gaussian perturbation with standard deviation σ. The results highlight the importance of initial conditions in determining learnability, particularly for almost-full parities and degree d parity. The study also contrasts its findings with those from statistical query (SQ) learning, which exhibits different behavior. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper looks at how well neural networks can learn certain types of patterns, called degree k parities. It finds that the way the network is set up to start with affects how well it learns these patterns. In particular, if the starting weights are chosen randomly from a special type of distribution, the network can learn almost-full parities and even the full parity itself. However, if the starting weights are chosen randomly but with a little extra noise added in, the network has trouble learning these same patterns. |
Keywords
» Artificial intelligence » Gradient descent
