DendroNN: Dendrocentric Neural Networks for Energy-Efficient Classification of Event-Based Data

Researchers have developed DendroNN, a novel neural network architecture inspired by brain dendrites that achieves up to 4x higher energy efficiency than current neuromorphic hardware for spatiotemporal event-based computing. The system uses spike sequence detection and a unique rewiring training method to process temporal data without requiring gradients or recurrent connections.

• →DendroNN introduces a dendrite-inspired neural network architecture that identifies spike sequences as spatiotemporal features for improved temporal processing.
• →The system achieves up to 4x higher efficiency than state-of-the-art neuromorphic hardware while maintaining comparable accuracy on audio classification tasks.
• →A novel rewiring training phase allows the network to learn non-differentiable spike sequences without using gradients.
• →The architecture eliminates the need for recurrence or delays that typically reduce hardware efficiency in spiking neural networks.
• →An asynchronous digital hardware design using time-wheel mechanisms enables event-driven processing with dynamic and static sparsity optimization.