Year: 2021
Type: conference-paper
Abstract: Language representations are known to carry certain associations (e.g., gendered connotations) which may lead to invalid and harmful predictions in downstream tasks.While existing methods are effective at mitigating such unwanted associations by linear projection, we argue that they are too aggressive: not only do they remove such associations, they also erase information that should be retained.To address this issue, we propose OS-CAR (Orthogonal Subspace Correction and Rectification), a balanced approach of mitigation that focuses on disentangling associations between concepts that are deemed problematic, instead of removing concepts wholesale.We develop new measurements for evaluating information retention relevant to the debiasing goal.Our experiments on genderoccupation associations show that OSCAR is a well-balanced approach that ensures that semantic information is retained in the embeddings and unwanted associations are also effectively mitigated.
Authors: Sunipa Dev, Tao Li, Jeff M Phillips, Vivek Srikumar
Institutions: University of California, Los Angeles, University of Utah
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
SDG: Reduced inequalities
Open Access status: hybrid
Awards: CAREER: Foundations for Geometric Analysis of Noisy Data, AI-DCL: Governing bias in AI system with humans in the decision loop, G5565683708, SaTC: CORE: Medium: Large-Scale Data Driven Anomaly Detection and Diagnosis from System Logs, III: Small: Persistent Data Summaries: Temporal Analytics on Big Data Histories, Computing Innovation Fellows 2020 Project, TWC: Medium: Collaborative: Seal: Secure Engine for AnaLytics - From Secure Similarity Search to Secure Data Analytics, NeTS: Medium: Collaborative Research: Detecting and Localizing Spectrum Offenders Using Crowdsourcing
OpenAlex ID: W3039559565