Summary of Scalable Bayesian Inference in the Era Of Deep Learning: From Gaussian Processes to Deep Neural Networks, by Javier Antoran
Scalable Bayesian Inference in the Era of Deep Learning: From Gaussian Processes to Deep Neural Networks
by Javier Antoran
First submitted to arxiv on: 29 Apr 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Machine Learning (cs.LG)
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 Medium Difficulty summary: This research paper proposes scalable methods to equip large neural networks with model uncertainty estimates. The authors develop techniques to leverage linearised Laplace approximation and conjugate Gaussian-linear models, allowing for Bayesian inference in neural networks. The approach involves using stochastic gradient descent (SGD) for posterior sampling in linear models and their convex duals: Gaussian processes. The method can be applied to large-scale neural networks like ResNet-50 trained on Imagenet, enabling the estimation of uncertainty for tasks such as 3d tomographic reconstructions. This work has implications for deep learning practices, particularly with regards to stochastic optimisation, early stopping, and normalisation layers. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Low Difficulty summary: This paper is about making big neural networks more reliable by adding a measure of how sure they are about their answers. Right now, these networks just give one answer without telling us how confident they are. This can lead to problems when we need to make multiple decisions based on the network’s predictions. The researchers developed new methods to calculate this uncertainty and applied it to large neural networks used for tasks like image recognition. This work could help us use deep learning models in more complex situations. |
Keywords
» Artificial intelligence » Bayesian inference » Deep learning » Early stopping » Resnet » Stochastic gradient descent
