Debiasing Pretrained Text Encoders by Paying Attention to Paying Attention
Recent studies in fair Representation Learning have observed a strong inclination for natural language processing (NLP) models to exhibit discriminatory stereotypes across gender, religion, race and many such social constructs. In comparison to the progress made in reducing bias from static word embeddings, fairness in sentence-level text encoders received little consideration despite their wider applicability in contemporary NLP tasks. In this paper, we propose a debiasing method for pre-trained text encoders that both reduces social stereotypes, and inflicts next to no semantic offset. Unlike previous studies that directly manipulate the embeddings, we suggest to dive deeper into the operation of these encoders, and pay more attention to the way they pay attention to different social groups. We find that the attention mechanism is the root of all stereotypes. Then, we work on model debiasing by redistributing the attention scores of a text encoder such that it forgets any preference to historically advantaged groups, and attends to all social classes with the same intensity. Our experiments confirm that we successfully reduce bias with little damage to semantic representation.
Code (0)
등록된 구현이 없습니다.
Tasks
FairnessRepresentation LearningSentenceWord EmbeddingsSimilar Papers 제목 키워드 기반
FairFil: Contrastive Neural Debiasing Method for Pretrained Text Encoders
Pretrained text encoders, such as BERT, have been applied increasingly in various natural language processing (NLP) tasks, and have recently demonstrated significant performance gains. However, recent studies have demons…
Contrastive LearningFairnessSentenceFairNVT: Improving Fairness via Noise Injection in Vision Transformers
This paper presents FairNVT, a lightweight debiasing framework for pretrained transformer-based encoders that improves both representation and prediction level fairness while preserving task accuracy. Unlike many existin…
FAIRT2V: Training-Free Debiasing for Text-to-Video Diffusion Models
Text-to-video (T2V) diffusion models have achieved rapid progress, yet their demographic biases, particularly gender bias, remain largely unexplored. We present FairT2V, a training-free debiasing framework for text-to-vi…
Text-to-Video GenerationFineDeb: A Debiasing Framework for Language Models
As language models are increasingly included in human-facing machine learning tools, bias against demographic subgroups has gained attention. We propose FineDeb, a two-phase debiasing framework for language models that s…
Language ModelingLanguage ModellingGeneral Phrase Debiaser: Debiasing Masked Language Models at a Multi-Token Level
The social biases and unwelcome stereotypes revealed by pretrained language models are becoming obstacles to their application. Compared to numerous debiasing methods targeting word level, there has been relatively less …