Learning the Truth From Only One Side of the Story
This addresses a fundamental issue in applications like lending and recommendation systems, offering a principled solution to mitigate sampling bias, though it is incremental as it builds on existing methods for a known bottleneck.
The paper tackles the problem of learning under one-sided feedback, where only positive predictions are observed, by focusing on generalized linear models and showing that unadjusted sampling bias leads to suboptimal convergence. It proposes an adaptive method with theoretical guarantees that empirically outperforms existing approaches, leveraging variance estimation for efficient learning under uncertainty.
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.