Comparative Analysis of Clustering and Projection Debiasing Methods for Gender Bias in Contextualized Embeddings

Abstract

Abstract: Taking an interdisciplinary approach to surveying issues around gender bias in textual and visual AI, we present literature on gender bias detection and mitigation in NLP, CV, as well as combined visual-linguistic models. We identify conceptual parallels between these strands of research as well as how methodologies were adapted cross-disciplinary from NLP to CV. We also find that there is a growing awareness for theoretical frameworks from the social sciences around gender in NLP that could be beneficial for aligning bias analytics in CV with human values and conceptualising gender beyond the

Research Question

Do clustering-based debiasing methods for contextualized embeddings demonstrate better downstream task performance on gender-bias benchmarks like WinoBias or CrowS-Pairs compared to projection-based methods?

Verification Level

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Truth-Engine Gate Verdict

This record has not completed Gate 2 of the verification pipeline (a type-checked Lean4 proof for mathematical claims, or a sealed-sandbox reproduction for empirical claims). It is a literature synthesis only. VERIFIED requires an attached reproducible artifact (Lean4 proof source, or repro script and results) before this status can be set; it is not derived from review score or claim count.

Quality Tier

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance