ε=2.0 provides the best privacy–utility balance, while ε=0.5 introduces excessive noise that degrades performance below baseline on ASSIST09¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? On ASSIST09, ε=2.0 retains a 14.19% improvement with 0.19% privacy cost, ε=1.0 retains 7.38%, and ε=0.5 yields −3.69% (worse than baseline). → 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 · i? · design · r2

Privacy–utility tradeoff analysis on ASSIST09 under local differential privacy (Table 3), where "ε= 2.0 setting provides the best balance"; stricter budgets degrade performance, with ε=0.5 introducing excessive noise (−3.69%).

"The ε= 2.0 setting provides the best balance: 14.19% improvement with only marginal degradation relative to the non-private model (0.19% privacy cost on ASSIST09)."

Discussion¶

Related Claims¶

• Laplace-based local differential privacy outperforms Gaussian noise under Rényi-DP accounting for bounded (0,1)K prediction outputs — related
• Heterogeneous multi-LLM federated inference improves cognitive diagnosis MAE over single-LLM baselines across three educational benchmarks with minimal privacy cost — related
• Residual correction is essential: removing it while retaining LDP yields MAE worse than the single-LLM baseline across all datasets — related
• Federated improvements over the single-LLM baseline are statistically significant across all three datasets, with Cohen's d from 0.62 to 0.89 — related
• PETs expand the value of data by enabling sharing where in-the-clear approaches are infeasible — related