Preface to ‘advancing uncertainty quantification in artificial intelligence systems using conformal prediction’

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract As artificial intelligence (AI) systems are being widely deployed in safety-critical and high-stakes applications (e.g. medical diagnosis, autonomous vehicles, financial risk assessment), there is a growing demand for providing reliable and trustworthy machine predictions. However, since AI models become more complex in structure (a prominent example being deep neural networks) and bigger in size (e.g. large language model systems such as ChatGPT and Gemini), being able to understand, explain and quantify confidence in their predictions are ongoing challenges. Therefore, this special issue is dedicated to exploring the forefront of reliable uncertainty quantification in AI systems, using conformal prediction (CP), a leading statistical framework that offers predictions with valid coverage guarantees under minimal assumptions. The issue comprises the most recent, most novel and most practical developments of CP-based methods in cutting-edge AI applications, highlighting improvements over traditional methods.