Learning with Minimal Human Feedback

Attend a lecture by Professor Alexandre Proutiere from KTH Royal Institute of Technology exploring the critical challenge of minimizing costly human feedback in machine learning tasks. Discover how to formalize this problem within the framework of online classification with expert guidance, particularly for question-answering applications. Learn about novel algorithms designed for both low- and high-budget scenarios, with theoretical foundations combining concentration-of-measure techniques and convex geometry tools. Examine regret upper bounds for these algorithms and explore their practical validation through experiments on real-world question-answering datasets using embeddings from state-of-the-art large language models. Gain insights into model alignment and fine-tuning strategies that reduce reliance on human intervention while maintaining performance. Understand the mathematical foundations underlying efficient learning systems that can operate with minimal expert guidance, making machine learning more scalable and cost-effective for practical applications.