A Multi-Core Controller for an Embedded AI System Supporting Parallel Recognition

Area of Science:

• Artificial Intelligence
• Computer Engineering
• Embedded Systems

Background:

• Centralized AI systems face network bandwidth and security challenges due to extensive data processing on servers.
• Federated learning offers a decentralized approach where clients train data locally, transmitting only results to a central server.
• This reduces network load and enhances data privacy by abstracting raw data.

Purpose of the Study:

• To propose a novel embedded AI system architecture for federated learning.
• To address the need for efficient resource utilization in resource-limited client systems.
• To design a flexible system composed of multiple embedded AI processors managed by a controller.

Main Methods:

• Designed a controller for arbitrating and utilizing multiple AI cores.
• Implemented the controller on a Field-Programmable Gate Array (FPGA).
• Validated the system's operation using image and speech processing applications and performance analysis via simulation.

Main Results:

• The proposed embedded AI system demonstrates flexible composition with various AI cores.
• The FPGA-implemented controller effectively manages multiple AI processors for federated learning tasks.
• Verified operational capabilities through practical image and speech application scenarios.

Conclusions:

• The developed embedded AI system provides a flexible and efficient solution for federated learning on resource-constrained devices.
• The novel controller design is crucial for managing distributed AI processing in embedded federated learning systems.
• This approach significantly improves network efficiency and data security in AI deployments.