Talk
Algorithms with (Distributional) Predictions
Talk Abstract
The field of algorithms with predictions has developed a large body of work showing how to use predictions to improve performance of algorithms. However, the vast majority of work in this space assumes that the prediction itself is non-probabilistic, even if it is generated by some stochastic process (such as a machine learning system). In contrast, many modern ML systems generate distributions over potential outcomes. In this talk we will cover some recent results using such “distributional” predictions.