Summary of Calibrated Computation-aware Gaussian Processes, by Disha Hegde et al.
Calibrated Computation-Aware Gaussian Processes
by Disha Hegde, Mohamed Adil, Jon Cockayne
First submitted to arxiv on: 11 Oct 2024
Categories
- Main: Machine Learning (stat.ML)
- Secondary: Machine Learning (cs.LG); Numerical Analysis (math.NA)
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 Gaussian processes are notorious for scaling cubically with the size of the training set, preventing application to very large regression problems. A new approach, Computation-aware Gaussian Processes (CAGPs), tackles this issue by exploiting probabilistic linear solvers to reduce complexity. However, current CAGP frameworks result in conservative uncertainty quantification, making the posterior unrealistic. This paper proves that a calibrated probabilistic linear solver induces a calibrated CAGP and proposes a new framework, CAGP-GS, using Gauss-Seidel iterations for the underlying solver. Compared to existing approaches, CAGP-GS performs well on low-dimensional test sets with few iterations. The calibratedness is tested on a synthetic problem, and performance is compared on a large-scale global temperature regression problem. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Gaussian processes are used in big data problems, but they get slower as the data grows. A new way to fix this problem uses special math tools called probabilistic linear solvers. This helps make predictions more realistic. But some existing methods don’t do this well. This research shows that if these math tools are used correctly, then the predictions will be better. The researchers also developed a new method, CAGP-GS, which is faster and better than other methods for certain problems. |
Keywords
* Artificial intelligence * Regression * Temperature
