FCPO: Federated Continual Policy Optimization for Real-Time High-Throughput Edge Video Analytics

2025-07-23

AI Summary

In edge video analytics (EVA), real-time, high-throughput scheduling across heterogeneous devices faces challenges including prolonged global decision cycles, poor scalability of local reinforcement learning (RL), and weak environmental adaptability. To address these, we propose the Federated Continual Policy Optimization (FCPO) framework, which integrates continual RL with federated RL. FCPO enables cross-device knowledge sharing and online environmental adaptation via personalized model aggregation and diversity-aware experience replay. It jointly optimizes batch size, input resolution, and multithreading policies. Evaluated on a real-world edge testbed, FCPO achieves over 5× higher system throughput, 60% lower end-to-end latency, 20% faster policy convergence, and up to 10× reduced memory overhead—significantly breaking the inherent trade-offs among convergence speed, generalization capability, and resource efficiency.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing
• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments

Abstract

The growing complexity of Edge Video Analytics (EVA) facilitates new kind of intelligent applications, but creates challenges in real-time inference serving systems. State-of-the-art (SOTA) scheduling systems optimize global workload distributions for heterogeneous devices but often suffer from extended scheduling cycles, leading to sub-optimal processing in rapidly changing Edge environments. Local Reinforcement Learning (RL) enables quick adjustments between cycles but faces scalability, knowledge integration, and adaptability issues. Thus, we propose FCPO, which combines Continual RL (CRL) with Federated RL (FRL) to address these challenges. This integration dynamically adjusts inference batch sizes, input resolutions, and multi-threading during pre- and post-processing. CRL allows agents to learn from changing Markov Decision Processes, capturing dynamic environmental variations, while FRL improves generalization and convergence speed by integrating experiences across inference models. FCPO combines these via an agent-specific aggregation scheme and a diversity-aware experience buffer. Experiments on a real-world EVA testbed showed over 5 times improvement in effective throughput, 60% reduced latency, and 20% faster convergence with up to 10 times less memory consumption compared to SOTA RL-based approaches.

Problem

• Optimize real-time Edge Video Analytics scheduling for dynamic environments
• Address scalability and adaptability issues in local Reinforcement Learning
• Improve throughput and latency in heterogeneous Edge device workloads

Innovation

• Combines Continual RL with Federated RL
• Dynamically adjusts batch sizes and resolutions
• Uses diversity-aware experience buffer