Summary of Towards Improved Variational Inference For Deep Bayesian Models, by Sebastian W. Ober
Towards Improved Variational Inference for Deep Bayesian Models
by Sebastian W. Ober
First submitted to arxiv on: 23 Jan 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 thesis, researchers explore ways to improve Bayesian deep learning by approximating the true posterior distribution using variational inference (VI). They propose a novel approach to VI that provides a unified view of inference for both Bayesian neural networks and deep Gaussian processes. The authors also demonstrate how VI can be improved in certain models by analytically removing symmetries from the posterior and performing inference on Gram matrices instead of features. This work has the potential to enable more effective use of VI in Bayesian learning, which could lead to better model selection and hyperparameter optimization. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Deep learning has made huge progress in many areas, but it often makes mistakes because it’s too sure of itself. To fix this, researchers try to put limits on how confident a deep model can be. They use something called Bayesian learning, where they guess what the model might look like and then adjust that guess based on new information. One way to do this is with something called variational inference (VI). This thesis looks at how VI works for deep models, especially neural networks and special kinds of neural networks called Gaussian processes. The authors also show how to make VI better in some cases by simplifying the math. Overall, this research could help us use VI more effectively, which would be really helpful for making decisions about what kind of model to use. |
Keywords
* Artificial intelligence * Deep learning * Hyperparameter * Inference * Optimization
