Toward a Better Understanding of Leaderboard

TL;DR

This paper analyzes the limitations of traditional leaderboards in machine learning competitions, proposes improvements to prevent overfitting, and discusses the theoretical aspects of leaderboard accuracy and complexity.

Contribution

It offers practical advice to improve leaderboard robustness, simplifies the Ladder leaderboard, and provides theoretical insights into sample complexity.

Findings

Ladder leaderboard can be simplified by removing redundant computations.

Sample complexity for accurate leaderboard estimation is cubic in the inverse of the precision.

Practical guidelines to prevent hacking and overfitting in leaderboards.

Abstract

The leaderboard in machine learning competitions is a tool to show the performance of various participants and to compare them. However, the leaderboard quickly becomes no longer accurate, due to hack or overfitting. This article gives two pieces of advice to prevent easy hack or overfitting. By following these advice, we reach the conclusion that something like the Ladder leaderboard introduced in [blum2015ladder] is inevitable. With this understanding, we naturally simplify Ladder by eliminating its redundant computation and explain how to choose the parameter and interpret it. We also prove that the sample complexity is cubic to the desired precision of the leaderboard.