Adversarial Removal of Demographic Attributes from Text Data

Abstract

Recent advances in Representation Learning and Adversarial Training seem to succeed in removing unwanted features from the learned representation. We show that demographic information of authors is encoded in -- and can be recovered from -- the intermediate representations learned by text-based neural classifiers. The implication is that decisions of classifiers trained on textual data are not agnostic to -- and likely condition on -- demographic attributes. When attempting to remove such demographic information using adversarial training, we find that while the adversarial component achieves chance-level development-set accuracy during training, a post-hoc classifier, trained on the encoded sentences from the first part, still manages to reach substantially higher classification accuracies on the same data. This behavior is consistent across several tasks, demographic properties and datasets. We explore several techniques to improve the effectiveness of the adversarial component. Our main conclusion is a cautionary one: do not rely on the adversarial training to achieve invariant representation to sensitive features.

References in corpus (8)

• Improving neural networks by preventing co-adaptation of feature detectors
• Semantics derived automatically from language corpora contain human-like biases
• Equality of Opportunity in Supervised Learning
• Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations
• Word Translation Without Parallel Data
• Adversarial Deep Averaging Networks for Cross-Lingual Sentiment Classification
• Controllable Invariance through Adversarial Feature Learning
• Towards Robust and Privacy-preserving Text Representations

Cited by in corpus (8)

• EDITS: Modeling and Mitigating Data Bias for Graph Neural Networks
• Hurtful Words: Quantifying Biases in Clinical Contextual Word Embeddings
• Modeling Techniques for Machine Learning Fairness: A Survey
• Exposing and Correcting the Gender Bias in Image Captioning Datasets and Models
• Online Abuse of UK MPs from 2015 to 2019: Working Paper
• Using Deep Networks for Scientific Discovery in Physiological Signals
• Fair Hate Speech Detection through Evaluation of Social Group Counterfactuals
• Transitioning from Real to Synthetic data: Quantifying the bias in model