Weighted Classification Cascades for Optimizing Discovery Significance in the HiggsML Challenge

TL;DR

This paper presents a new iterative method for optimizing discovery significance in high energy physics, using weighted classification and convex duality, validated on the HiggsML challenge data.

Contribution

It introduces a novel minorization-maximization algorithm that links weighted classification error improvements to discovery significance enhancement.

Findings

Effective optimization of discovery significance demonstrated

Algorithm outperforms baseline methods in HiggsML challenge

Theoretical guarantees connect classification error reduction to significance increase

Abstract

We introduce a minorization-maximization approach to optimizing common measures of discovery significance in high energy physics. The approach alternates between solving a weighted binary classification problem and updating class weights in a simple, closed-form manner. Moreover, an argument based on convex duality shows that an improvement in weighted classification error on any round yields a commensurate improvement in discovery significance. We complement our derivation with experimental results from the 2014 Higgs boson machine learning challenge.