Privacy-preserving heterogeneous multi-LLM federated inference framework for cognitive diagnosis

Research Method · All research methods
Evidence · 3 claims (3 for) · 1 study (1 design), q2 · 0 of 1 report an effect size · 3 claims rest on one study

Description

A four-stage framework in which heterogeneous commercial LLMs (LLaMA-3.3-70B, GPT-4o-mini, Claude-3-Haiku) independently predict student knowledge states via engineered prompts, add calibrated Laplace noise locally to their prediction outputs, and aggregate with residual correction. The article describes it as "the first privacy-preserving heterogeneous multi-LLM federated inference framework for cognitive diagnosis", operating entirely at the prediction level with models frozen and accessible only via black-box APIs. It integrates Q-matrix theory to constrain prompts to relevant concepts and rests on an honest-but-curious trust paradigm.

Accounts

• Privacy-preserving heterogeneous multi-LLM federated inference framework for cognitive diagnosis: A four-stage framework in which heterogeneous commercial LLMs (LLaMA-3.3-70B, GPT-4o-mini, Claude-3-Haiku) independently predict student knowledge states via engineered prompts, add calibrated Laplace noise locally to their prediction outputs, and aggregate with residual correction. The article describes it as "the first privacy-preserving heterogeneous multi-LLM federated inference framework for cognitive diagnosis", operating entirely at the prediction level with models frozen and accessible only via black-box APIs. It integrates Q-matrix theory to constrain prompts to relevant concepts and rests on an honest-but-curious trust paradigm. (Yagna Manasa Boyapati et al. (2026))

Claims

• Heterogeneous multi-LLM federated inference improves cognitive diagnosis MAE over single-LLM baselines across three educational benchmarks with minimal privacy cost [+M]
• Residual correction is essential: removing it while retaining LDP yields MAE worse than the single-LLM baseline across all datasets [+M]
• ε=2.0 provides the best privacy–utility balance, while ε=0.5 introduces excessive noise that degrades performance below baseline on ASSIST09 [+M]

Related Research Methods

-

Key Sources

• 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

Page metadata

Field	Value
type	research-method
id	privacy-preserving-heterogeneous-multi-llm-federated-inference-framework-for-cognitive-dia
title	Privacy-preserving heterogeneous multi-LLM federated inference framework for cognitive diagnosis
description	A four-stage framework in which heterogeneous commercial LLMs (LLaMA-3.3-70B, GPT-4o-mini, Claude-3-Haiku) independently predict student knowledge states via engineered prompts, add calibrated Laplace noise locally to their prediction outputs, and aggregate with residual correction.
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content

Frontmatter is the page's machine-readable record — see the Schema & Guide.