Summary of Training-time Neuron Alignment Through Permutation Subspace For Improving Linear Mode Connectivity and Model Fusion, by Zexi Li et al.
Training-time Neuron Alignment through Permutation Subspace for Improving Linear Mode Connectivity and Model Fusion
by Zexi Li, Zhiqi Li, Jie Lin, Tao Shen, Tao Lin, Chao Wu
First submitted to arxiv on: 2 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 In this paper, researchers tackle a long-standing issue in deep learning: the scattered nature of weight space solutions despite identical initializations. They show that permutation symmetry can reduce these barriers post-training but this method is less effective for complex models. To overcome this challenge, they propose training-time neuron alignment (TNA) as an alternative approach. TNA involves pruning at initialization and introducing a partial gradient mask during training to reduce the Linear Mode Connectivity (LMC) landscape barriers. The authors validate their theory through experiments and demonstrate the effectiveness of TNA in wide model fusion applications, particularly in federated learning. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In this study, scientists explore how to improve deep learning models by reducing barriers in the weight space. They found that these barriers happen even when starting with the same initial weights. To solve this problem, they came up with a new method called training-time neuron alignment (TNA). This approach helps reduce these barriers without needing extra computations. The researchers tested TNA and showed it works well for combining different models together, which is useful in situations where data is shared between different groups. |
Keywords
* Artificial intelligence * Alignment * Deep learning * Federated learning * Mask * Pruning
