Modeling and Simultaneously Removing Bias via Adversarial Neural Networks

TL;DR

This paper introduces an adversarial neural network approach to model and remove bias in machine learning predictions, demonstrated on paid search auction data to produce bias-invariant representations.

Contribution

The paper proposes a novel adversarial neural network method to create bias-invariant data representations, addressing bias in training data caused by deployed models.

Findings

Effective bias removal demonstrated on synthetic data.

Successful application to real-world paid search auction data.

Produces bias-invariant representations improving model fairness.

Abstract

In real world systems, the predictions of deployed Machine Learned models affect the training data available to build subsequent models. This introduces a bias in the training data that needs to be addressed. Existing solutions to this problem attempt to resolve the problem by either casting this in the reinforcement learning framework or by quantifying the bias and re-weighting the loss functions. In this work, we develop a novel Adversarial Neural Network (ANN) model, an alternative approach which creates a representation of the data that is invariant to the bias. We take the Paid Search auction as our working example and ad display position features as the confounding features for this setting. We show the success of this approach empirically on both synthetic data as well as real world paid search auction data from a major search engine.