Learning the Truth From Only One Side of the Story

TL;DR

This paper addresses the challenge of learning from one-sided feedback in machine learning, proposing an adaptive method with theoretical guarantees that improves convergence and outperforms existing techniques.

Contribution

The authors introduce an adaptive approach with variance estimation for learning from biased data, providing theoretical guarantees and empirical improvements.

Findings

The method converges more reliably than previous approaches.

It outperforms existing methods in empirical evaluations.

The approach effectively mitigates sampling bias effects.

Abstract

Learning under one-sided feedback (i.e., where we only observe the labels for examples we predicted positively on) is a fundamental problem in machine learning -- applications include lending and recommendation systems. Despite this, there has been surprisingly little progress made in ways to mitigate the effects of the sampling bias that arises. We focus on generalized linear models and show that without adjusting for this sampling bias, the model may converge suboptimally or even fail to converge to the optimal solution. We propose an adaptive approach that comes with theoretical guarantees and show that it outperforms several existing methods empirically. Our method leverages variance estimation techniques to efficiently learn under uncertainty, offering a more principled alternative compared to existing approaches.