Pulse-Driven Neural Architecture: Learnable Oscillatory Dynamics for Robust Continuous-Time Sequence Processing

Researchers introduce PDNA (Pulse-Driven Neural Architecture), a new continuous-time neural network that incorporates learnable oscillatory dynamics to improve robustness when input sequences are interrupted. The method shows significant performance improvements on sequential MNIST tasks, with the pulse variant achieving a 4.62 percentage point advantage over baseline models.

• →PDNA augments continuous-time recurrent networks with learnable oscillatory dynamics that maintain state evolution independently of external input.
• →The architecture includes a pulse module generating structured oscillations and a self-attend module applying recurrent self-attention.
• →Testing on sequential MNIST showed statistically significant improvements in gap robustness when input sequences are interrupted.
• →The pulse variant demonstrated a 4.62 percentage point advantage with large effect size over baseline models.
• →Results suggest biologically-inspired oscillatory mechanisms can enhance temporal robustness in AI models.