← ML Research Wiki / 2407.17712

Improving Online Algorithms via ML Predictions *

Ravi Kumar [email protected], Manish Purohit [email protected], Zoya Svitkina (2024)

Paper Information
arXiv ID
Venue
Neural Information Processing Systems
Domain
Computer Science
Code
Available
Reproducibility
5/10

Abstract

In this work we study the problem of using machine-learned predictions to improve the performance of online algorithms.We consider two classical problems, ski rental and non-clairvoyant job scheduling, and obtain new online algorithms that use predictions to make their decisions.These algorithms are oblivious to the performance of the predictor, improve with better predictions, but do not degrade much if the predictions are poor.

Summary

This paper investigates enhancing the performance of online algorithms through machine-learned (ML) predictions in two established problems: ski rental and non-clairvoyant job scheduling. The authors propose algorithms that are adaptive to the quality of predictions, maintaining robust performance with poor predictions while achieving optimal performance with good predictions. They introduce notions of robustness and consistency related to competitive ratios against optimal offline algorithms. The study includes a deterministic and a randomized algorithm for the ski rental problem, illustrating the use of predictions carefully to ensure robustness. For non-clairvoyant scheduling, they develop a preferential round-robin algorithm combining aspects of predicted processing times and traditional scheduling, allowing better performance than classical methods under good predictions. Experimental results indicate the algorithms significantly outperform traditional approaches with varying prediction errors.

Methods

This paper employs the following methods:

  • ML predictions
  • online algorithms
  • deterministic algorithms
  • randomized algorithms

Models Used

  • None specified

Datasets

The following datasets were used in this research:

  • None specified

Evaluation Metrics

  • None specified

Results

  • New online algorithms for ski rental and non-clairvoyant job scheduling
  • Algorithms demonstrate improved robustness and consistency with predictions
  • Experimental results show significant performance improvements over classical algorithms

Limitations

The authors identified the following limitations:

  • Careful prediction usage is essential for maintaining robustness. Predictions' performance affects adaptability of the algorithms.

Technical Requirements

  • Number of GPUs: None specified
  • GPU Type: None specified

Keywords

Online algorithms Machine learning predictions Competitive analysis Scheduling Ski rental

Papers Using Similar Methods

External Resources