Trade-offs in integrating TinyML and Real-time ML with 6G connectivity

Characterize the trade-offs between energy efficiency and inference/training accuracy arising from the integration of Tiny Machine Learning and Real-time Machine Learning with 6G wireless connectivity for distributed, on-device training and inference in mobile artificial intelligence applications.

Background

The paper argues that centralized learning is ill-suited for real-time control and adaptation in latency-constrained wireless environments, motivating a shift to distributed learning across non-homogeneous devices under strict communication and energy constraints.

Within this context, the authors emphasize that combining TinyML and Real-time ML with 6G connectivity is not yet well-understood, especially in terms of balancing energy consumption with model accuracy during both training and inference across devices.