Randomized trial evaluates an end-to-end event-driven system, suggesting a pathway for low-power learning.
Key Points
This research aims to develop a hardware-oriented framework for integrating on-device learning in autonomous systems to enhance energy efficiency and reduce latency.
Introduced a hierarchical spiking predictive coding framework with implicit prediction error encoding.
Utilized local lateral and supervisory feedback connections to improve computational efficiency.
Evaluated the system on neuromorphic datasets under real-time hardware constraints with a fixed temporal resolution.
The SPC framework effectively identifies stimuli from transient event streams during on-device learning.
Achieved stable learning performance with low power consumption in resource-constrained environments.