Privacy-Preserving Federated Action Recognition via Differentially Private Selective Tuning and Efficient Communication

Federated video action recognition enables collaborative model training without sharing raw video data, yet remains vulnerable to two key challenges: $\mathit{\text{model}}\mathit{\text{exposure}}$ and $\mathit{\text{communication}}\mathit{\text{overhead}}$. Gradients exchanged between clients and the server can leak private motion patterns, while full-model synchronization of high-dimensional video networks causes significant bandwidth and communication costs. To address these issues, we propose $\mathit{\text{Federated}}\mathit{\text{Differential}}\mathit{\text{Privacy}}\mathit{\text{with}}\mathit{\text{Selective}}\mathit{\text{Tuning}}\mathit{\text{and}}\mathit{\text{Efficient}}\mathit{\text{Communication}}\mathit{\text{for}}\mathit{\text{Action}}\mathit{\text{Recognition}}$, namely $\mathit{\text{FedDP-STECAR}}$. Our $\mathit{\text{FedDP-STECAR}}$ framework selectively fine-tunes and perturbs only a small subset of task-relevant layers under Differential Privacy (DP), reducing the surface of information leakage while preserving temporal coherence in video features. By transmitting only the tuned layers during aggregation, communication traffic is reduced by over 99% compared to full-model updates. Experiments on the UCF-101 dataset using the MViT-B-16x4 transformer show that $\mathit{\text{FedDP-STECAR}}$ achieves up to $\mathbf{\text{70.2\%}}\mathbf{\text{higher}}\mathbf{\text{accuracy}}$ under strict privacy ($\epsilon =0.65$) in centralized settings and $\mathbf{\text{48\%}}\mathbf{\text{faster}}\mathbf{\text{training}}$ with $\mathbf{\text{73.1\%}}\mathbf{\text{accuracy}}$ in federated setups, enabling scalable and privacy-preserving video action recognition. Code available at https://github.com/izakariyya/mvit-federated-videodp