Lion Optimizer's Impact on ModernBERT Cross-Encoder Robustness Against Domain Shift

Abstract

Abstract: Language models have revolutionized natural language processing, becoming an integral part of many applications. However, these models often exhibit societal biases embedded in their training data, raising concerns about their fairness and ethical deployment. Measuring these biases usually requires creating datasets with time-consuming human annotation, which is costly and hard to expand. To address this challenge, we propose a data curation framework and CineBias, a novel dataset of 1,012 stereotypical sentence pairs covering seven bias categories, extracted from Hollywood movie subtitles wit

Research Question

Does the Lion optimizer improve the robustness of ModernBERT cross-encoders against domain shift when evaluated on out-of-distribution retrieval benchmarks beyond MS MARCO?

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