Summary of Fsp-laplace: Function-space Priors For the Laplace Approximation in Bayesian Deep Learning, by Tristan Cinquin et al.
FSP-Laplace: Function-Space Priors for the Laplace Approximation in Bayesian Deep Learning
by Tristan Cinquin, Marvin Pförtner, Vincent Fortuin, Philipp Hennig, Robert Bamler
First submitted to arxiv on: 18 Jul 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 paper proposes a novel approach to epistemic uncertainty estimation in deep networks by directly placing a prior on function space. This addresses limitations of Laplace approximations with isotropic Gaussian priors, which can cause pathological behavior as depth increases. The method, inspired by Gaussian process (GP) priors, allows for structured and interpretable inductive biases, such as regularity or periodicity, in function space. By leveraging scalable methods from matrix-free linear algebra, the approach provides improved results where prior knowledge is abundant, while staying competitive for black-box supervised learning problems. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A team of researchers has developed a new way to estimate uncertainty in deep neural networks. This is important because it allows us to better understand how certain we are about our predictions. The current method uses something called Laplace approximations, which can be effective but have some limitations. To solve these issues, the team came up with a new approach that involves placing a prior on function space. This means they’re putting in place a kind of “prior knowledge” to help guide their training process. The result is a method that works well when we have prior knowledge about what’s likely to happen, and it’s also competitive with other methods for problems where we don’t have much prior knowledge. |
Keywords
* Artificial intelligence * Supervised
