Summary of Bayesian Deep Learning Via Expectation Maximization and Turbo Deep Approximate Message Passing, by Wei Xu et al.
Bayesian Deep Learning Via Expectation Maximization and Turbo Deep Approximate Message Passing
by Wei Xu, An Liu, Yiting Zhang, Vincent Lau
First submitted to arxiv on: 12 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI)
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 proposed algorithm, EM-TDAMP, tackles the limitations of traditional deep learning methods by formulating DNN learning and compression as a sparse Bayesian inference problem. This approach employs group sparse priors for structured model compression, overcoming drawbacks of SGD-based learning algorithms and regularization-based model compression methods. By leveraging an expectation-maximization framework with a turbo deep approximate message passing algorithm, EM-TDAMP efficiently estimates posterior distributions for parameters and updates hyperparameters. The authors also extend this approach to develop a Bayesian federated learning framework, which enables clients to perform local computations and accelerates global convergence. Empirical evaluations demonstrate the effectiveness of EM-TDAMP in Boston housing price prediction and handwriting recognition tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary EM-TDAMP is an innovative algorithm that helps deep neural networks learn better and use less storage space. It works by treating model compression as a math problem, using ideas from statistics to find the best way to compress models while keeping them accurate. The algorithm uses something called expectation maximization to figure out how to update the model’s parameters and hyperparameters. This allows it to be more efficient and scalable for big datasets. The authors also show how this approach can be used in a special type of machine learning, called federated learning, which lets multiple devices work together to train a single model. |
Keywords
* Artificial intelligence * Bayesian inference * Deep learning * Federated learning * Machine learning * Model compression * Regularization
