Summary of Training Bayesian Neural Networks with Sparse Subspace Variational Inference, by Junbo Li et al.
Training Bayesian Neural Networks with Sparse Subspace Variational Inference
by Junbo Li, Zichen Miao, Qiang Qiu, Ruqi Zhang
First submitted to arxiv on: 16 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 This paper presents a novel approach called Sparse Subspace Variational Inference (SSVI) that efficiently trains Bayesian neural networks (BNNs) while maintaining high sparsity throughout the training and inference phases. SSVI addresses the challenge of substantially increased training costs by introducing a removal-and-addition strategy to optimize the sparse subspace basis selection, guided by novel criteria based on weight distribution statistics. The approach achieves impressive results, including up to 20x FLOPs reduction during training compared to dense VI training and 10-20x compression in model size with minimal performance drop. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to train special kinds of neural networks called Bayesian neural networks (BNNs). These networks are great at telling us how sure they are about their predictions, but they can be very slow and use a lot of computer power. The researchers came up with an idea called Sparse Subspace Variational Inference (SSVI) that helps train BNNs more quickly and uses less energy. They tested SSVI on some big problems and found that it can reduce the amount of computer power needed by 10-20 times while still getting good results. |
Keywords
* Artificial intelligence * Inference
