Summary of How Do the Architecture and Optimizer Affect Representation Learning? on the Training Dynamics Of Representations in Deep Neural Networks, by Yuval Sharon and Yehuda Dar
How Do the Architecture and Optimizer Affect Representation Learning? On the Training Dynamics of Representations in Deep Neural Networks
by Yuval Sharon, Yehuda Dar
First submitted to arxiv on: 27 May 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 explores how representations in deep neural networks evolve during training, focusing on overparameterized settings where the model continues to learn beyond perfecting its training data. The authors examine the similarity of DNN layer representations throughout training using centered kernel alignment (CKA) and decision region similarities. Visualization and analysis reveal patterns in training dynamics, including memorization phases, depending on layer depth, architecture, and optimizer. For instance, SGD training shows more distinct phases than Adam training, while Vision Transformer (ViT) layers exhibit synchronized representation learning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper looks at how deep neural networks learn during training. It wants to know what happens in the network as it gets better at doing its job. The authors use special tools to compare the different parts of the network and see how they change over time. They find that the way the network learns changes depending on things like the type of data it’s learning from, the method used to update the weights, and the design of the network itself. |
Keywords
» Artificial intelligence » Alignment » Representation learning » Vision transformer » Vit
