Federated improvements over the single-LLM baseline are statistically significant across all three datasets, with Cohen's d from 0.62 to 0.89¶

Claim · All claims
Evidence · 1 study · design r2 · q2 · i3 large
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i3 Paired t-tests show p<0.001 with Cohen's d=0.85 for ε=2.0 on ASSIST09, p<0.01 with d=0.62 on GSM8K, and p<0.001 with d=0.78 on UCI; the non-private ASSIST09 result shows d=0.89. → Yagna Manasa Boyapati 2026

Evidence¶

Yagna Manasa Boyapati 2026¶

Yagna Manasa Boyapati, Chong Yu, Tianyu Jiang, Justin Zhan. (2026). Privacy-Preserving Heterogeneous Multi-LLM Federated Inference for Cognitive Diagnosis. arXiv preprint. https://arxiv.org/abs/2609.02947 doi:10.48550/arxiv.2609.02947

q2 · i3 · design · r2

Statistical significance analysis (Appendix C) using paired t-tests; ASSIST09 non-private federation showed p<0.001 with Cohen's d=0.89, and similar significance was observed across GSM8K (p<0.01, d=0.62) and UCI (p<0.001, d=0.78).

"For ε= 2.0 (MAE 0.2068 versus baseline 0.2410), significance re- mains strong atp<0.001 with Cohen’sd= 0.85 , meaning that a moderate level of differential pri- vacy ensures the preservation of statistical signif- icance"

Discussion¶

Related Claims¶

• Heterogeneous multi-LLM federated inference improves cognitive diagnosis MAE over single-LLM baselines across three educational benchmarks with minimal privacy cost — a broader claim this one bears on
• ε=2.0 provides the best privacy–utility balance, while ε=0.5 introduces excessive noise that degrades performance below baseline on ASSIST09 — related
• Improvements over baselines are statistically significant, with Cohen's d ranging from 1.2 to 1.8 for key metrics — related
• Residual correction is essential: removing it while retaining LDP yields MAE worse than the single-LLM baseline across all datasets — related