Table 10 Sensitivity analysis for the contrastive loss weight λ
From: Mitigating the risk of health inequity exacerbated by large language models
λ | NDCG@10 (CTM) | Error rate (MQA) | Fairness improvement (EO diff) |
|---|---|---|---|
0.05 | 70.1 ± 0.6% | 20.5 ± 0.7% | 30% |
0.10 | 70.3 ± 0.5% | 19.8 ± 0.5% | 35% |
0.20 | 70.0 ± 0.7% | 20.2 ± 0.6% | 32% |
