arXiv subjects

Le Huy Khiem

Publications and source records attributed to Le Huy Khiem.

3 recordsLinked to original sources

arXiv2026

Onboarding Without Forgetting: Hypernetwork Personalization with Data-Free Replay for Personalized Federated Learning

Federated Learning (FL) enables collaborative training across distributed clients without sharing raw data, offering strong privacy benefits. However, most methods assume all clients remain available throughout training, which is unrealistic as new clients often join over time. We study this setting, where the task and label space stay fixed but clients arrive in batches. Our analysis reveals two key challenges: updating the shared model only with new clients harms existing clients, while freezing it protects them but blocks gains from new knowledge. To capture these trade-offs, we introduce Proactive Adaptation (PA) for onboarding gains and Retroactive Improvement (RI) for changes in earlier clients without retraining. We then propose pFedDSH, which combines a central hypernetwork for personalized initialization, batch-specific binary masks for capacity preservation and allocation, and server-side data-free replay to propagate improvements without exposing client data. Experiments show that pFedDSH preserves stability for existing clients while keeping communication and adaptation costs unchanged for new clients.

cs.LG↗

arXiv2026

Building Scaffolding Dialogue Data with LLM-Simulated Novices

High-quality, multi-turn instructional dialogues between novices and experts are essential for developing AI systems that support teaching, learning, and decision-making. These dialogues often involve scaffolding -- the process by which an expert supports a novice's thinking through questions, feedback, and step-by-step guidance. However, such data are scarce due to privacy concerns in recording and the vulnerability inherent in help-seeking. We present SimInstruct, a scalable, expert-in-the-loop tool for collecting scaffolding dialogues. Using teaching development coaching as an example domain, SimInstruct simulates novice instructors via LLMs, varying their teaching challenges and LLM's persona traits, while human experts provide multi-turn feedback, reasoning, and instructional support. This design enables the creation of realistic, pedagogically rich dialogues without requiring real novice participants. Our results reveal that persona traits, such as extroversion and introversion, meaningfully influence how experts engage. Compared to real mentoring recordings, SimInstruct dialogues demonstrate comparable pedagogical relevance and cognitive depth. Experts also reported the process as engaging and reflective, improving both data quality and their own professional insight. We further fine-tuned a LLaMA model to be an expert model using the augmented dataset, which outperformed GPT-4o in instructional quality. Our analysis highlights GPT-4o's limitations in weak reflective questioning, overuse of generic praise, a condescending tone, and a tendency to overwhelm novices with excessive suggestions.

cs.AI↗

arXiv2026

Automated Benchmark Generation from Domain Guidelines Informed by Bloom's Taxonomy

Open-ended question answering (QA) evaluates a model's ability to perform contextualized reasoning beyond factual recall. This challenge is especially acute in practice-based domains, where knowledge is procedural and grounded in professional judgment, while most existing LLM benchmarks depend on pre-existing human exam datasets that are often unavailable in such settings. We introduce a framework for automated benchmark generation from expert-authored guidelines informed by Bloom's Taxonomy. It converts expert practices into implicit violation-based scenarios and expands them into auto-graded multiple-choice questions (MCQs) and multi-turn dialogues across four cognitive levels, enabling deterministic, reproducible, and scalable evaluation. Applied to three applied domains: teaching, dietetics, and caregiving, we find differences between model and human-like reasoning: LLMs sometimes perform relatively better on higher-order reasoning (Analyze) but fail more frequently on lower-level items (Remember). We produce large-scale, psychometrically informed benchmarks that surface these non-intuitive model behaviors and enable evaluation of contextualized reasoning in real-world settings.

cs.CL↗