Summary of Learning Associative Memories with Gradient Descent, by Vivien Cabannes et al.
Learning Associative Memories with Gradient Descent
by Vivien Cabannes, Berfin Simsek, Alberto Bietti
First submitted to arxiv on: 28 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); 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 the training dynamics of an associative memory module that stores outer products of token embeddings. By reducing this problem to a system of particles interacting according to data distribution and correlations between embeddings, the authors provide insights into overparameterized and underparameterized regimes. They find logarithmic growth of classification margins in the former and oscillatory transitory regimes due to correlated embeddings and memory interferences. Large step sizes can create benign loss spikes but accelerate convergence. The cross-entropy loss leads to suboptimal memorization schemes in underparameterized regimes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how a special kind of computer memory works when it’s trained on words and their meanings. They turn this problem into an analogy about particles interacting with each other, which helps them understand what happens during the training process. They find that when there’s too much information stored, the memory gets stuck in patterns that aren’t very useful. But they also show that if the learning rate is too high, the memory will learn too slowly and might even forget some things. |
Keywords
* Artificial intelligence * Classification * Cross entropy * Token
