Embedding artificial intelligence in retail

This talk provides an introduction to problems of real-time repeated decision-making in the face of uncertainty, commonly modelled via ‘multi-armed bandits’. Such problems arise in advertising (choosing what products to recommend), resource and inventory management (choosing what combinations of stock to hold), and the tuning of machine learning algorithms (choosing which parameters or frameworks to use). We covered some key models and algorithmic approaches to such problems which address the particular challenge of balancing between exploring the quality of poorly understood actions (products, stock levels, parameters) and exploiting the strong performance of well understood actions.