Summary of Posterior Uncertainty Quantification in Neural Networks Using Data Augmentation, by Luhuan Wu et al.
Posterior Uncertainty Quantification in Neural Networks using Data Augmentation
by Luhuan Wu, Sinead Williamson
First submitted to arxiv on: 18 Mar 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 This paper addresses the problem of uncertainty quantification in deep learning by introducing a predictive framework that captures model parameter uncertainty. The authors argue that traditional deep ensembling methods are mis-specified since they assume future data is supported only on existing observations, which is rarely true. To overcome this limitation, they propose MixupMP, a method that uses popular data augmentation techniques to construct a more realistic predictive distribution. MixupMP operates as a drop-in replacement for deep ensembles and returns samples from an implicitly defined Bayesian posterior. Empirical analysis shows that MixupMP achieves better predictive performance and uncertainty quantification on various image classification datasets compared to existing Bayesian and non-Bayesian approaches. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how to make predictions in artificial intelligence more reliable by accounting for the unknowns. Right now, deep learning models are great at recognizing things like images or speech, but they don’t always tell us how sure they are. The authors show that traditional methods are not very good at this and propose a new way called MixupMP that uses random simulations to make predictions more accurate. They test it on some image classification tasks and find that it works better than existing methods. |
Keywords
* Artificial intelligence * Data augmentation * Deep learning * Image classification
