Stop Guessing, Start Discovering Trade-offs in Your ML Models скачать в хорошем качестве

Stop Guessing, Start Discovering Trade-offs in Your ML Models

Title:Stop Guessing, Start Discovering Trade-offs in Your ML Models Speaker: Andrei Paleyes (https://paleyes.info/) Slides: https://paleyes.info/talks/2026_01_15... Abstract: The process of deploying a machine learning model for real-file decision making is rarely about prediction accuracy alone. It is often necessary to consider additional requirements that impact cost of operation, regulatory compliance, explainability, and more. Yet we often leave these important aspects as an afterthought, setting ad-hoc constraints and arbitrary thresholds around the learning process. This talk argues that we should stop guessing and start discovering these trade-offs. We will show how to think about such situations as multi-objective optimisation problems, and use Bayesian optimisation to solve them. Through concrete case studies spanning privacy, fairness, and energy efficiency, we will see how one approach can be utilised efficiently in different settings, improving upon existing practices or discovering new insights along the way. By the end of the talk listeners will be equipped with an intuitive understanding of how to apply same techniques to domains and objectives that is most relevant to them.