Algorithms with Predictions

TL;DR

This paper presents algorithms that leverage machine learning predictions to improve performance in algorithms, achieving near-optimal results with good predictions and fallback to worst-case behavior when predictions are inaccurate.

Contribution

It introduces a framework for algorithms that adaptively use machine learning predictions to enhance efficiency while maintaining robustness against prediction errors.

Findings

Algorithms perform near optimally with accurate predictions.

They revert to worst-case guarantees when predictions are poor.

The approach bridges machine learning and traditional algorithm analysis.

Abstract

We introduce algorithms that use predictions from machine learning applied to the input to circumvent worst-case analysis. We aim for algorithms that have near optimal performance when these predictions are good, but recover the prediction-less worst case behavior when the predictions have large errors.