Summary of Towards a Framework For Deep Learning Certification in Safety-critical Applications Using Inherently Safe Design and Run-time Error Detection, by Romeo Valentin
Towards a Framework for Deep Learning Certification in Safety-Critical Applications Using Inherently Safe Design and Run-Time Error Detection
by Romeo Valentin
First submitted to arxiv on: 12 Mar 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 This paper addresses the critical issue of certifying deep learning-based systems for deployment in safety-critical applications, such as aviation. The authors investigate methodologies from machine learning research aimed at verifying robustness and reliability, evaluating their applicability to real-world problems. A novel framework is established, comprising inherently safe design and run-time error detection. The paper showcases a concrete use case from aviation, demonstrating how deep learning models can recover disentangled variables through weakly-supervised representation learning. This design is less prone to common model failures and can be verified to encode underlying mechanisms governing data. The authors also investigate four techniques for run-time safety: uncertainty quantification, out-of-distribution detection, feature collapse, and adversarial attacks. A set of desiderata is formulated for a certified model, and a novel model structure is proposed that meets these requirements, making regression and uncertainty predictions while detecting out-of-distribution inputs without requiring regression labels. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper talks about how to make sure deep learning models are safe and reliable enough to be used in important situations like flying an airplane. Right now, there’s no way to officially check if a model is good enough for these situations. The authors explore different ways to do this, including making the model “inherently” safer and checking it while it’s running. They use a real-life example from aviation to show how their ideas work. They also discuss some other techniques that can help keep the model safe, like detecting when something is going wrong or when data doesn’t match what the model expects. The authors propose a new way of building models that meets these safety standards and can even predict uncertainty. |
Keywords
* Artificial intelligence * Deep learning * Machine learning * Regression * Representation learning * Supervised
