Summary of On the Calibration Of Epistemic Uncertainty: Principles, Paradoxes and Conflictual Loss, by Mohammed Fellaji et al.
On the Calibration of Epistemic Uncertainty: Principles, Paradoxes and Conflictual Loss
by Mohammed Fellaji, Frédéric Pennerath, Brieuc Conan-Guez, Miguel Couceiro
First submitted to arxiv on: 16 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 explores the concept of epistemic uncertainty in deep learning models such as Deep Ensembles, Bayesian Deep Networks, and Evidential Deep Networks. Epistemic uncertainty is a measure of how much a model’s predictions are affected by its own internal workings rather than external factors. The authors find that traditional methods for estimating epistemic uncertainty often fail to meet two fundamental requirements: decreasing with more training data and increasing with more expressive models. This discrepancy raises questions about the usefulness of these measures. To address this issue, the authors propose a regularization function called conflictual loss, which aims to restore the expected behavior of epistemic uncertainty while maintaining model performance. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper is about how some deep learning models are unsure about their predictions because they don’t always agree with each other. This “epistemic uncertainty” is important because it shows when a model is making a guess rather than being sure. However, the authors found that most methods for measuring epistemic uncertainty don’t follow two basic rules: getting better as more data is used and increasing as models become more complex. This is weird! The paper proposes a new way to measure epistemic uncertainty that follows these rules, making it more useful. |
Keywords
» Artificial intelligence » Deep learning » Regularization
