submitted2025-present
Federated Deep Equilibrium Learning over Resource-Constrained Edge Networks
Federated deep equilibrium models for heterogeneous clients where memory, energy, data, and communication capacity vary across the network.
NLP + visionevaluation domains
A model that solves for equilibrium
Deep equilibrium models represent an effectively infinite-depth network through a fixed point. FeDEQ asks how that formulation can be trained across clients with non-IID data and unequal resource budgets.
My work covers implementation, aggregation studies, and experiments spanning NLP and vision. The engineering emphasis is reproducibility: controlled data partitions, explicit communication accounting, stable fixed-point solvers, and comparable baselines.