Tag
The paper proposes FedTAR, a task-aware federated fine-tuning method for MoE-based large language models that aligns local updates with task preferences to preserve expert specialization and improve performance under heterogeneous data.
Setoka is a benchmark for evaluating memory-augmented personalized agents' ability to understand users hierarchically (semantic memory, episodic memory, behavior patterns, personality traits) from heterogeneous data, revealing that current memory systems struggle with tasks requiring cross-source integration and abstraction.
This paper introduces FedeKD, a reliability-aware framework for federated knowledge distillation that uses an energy-based gating mechanism to mitigate negative transfer in heterogeneous settings. The authors demonstrate that weighting knowledge transfer based on sample-wise trust improves robustness and predictive performance without requiring public datasets.