TECHNOLOGY
Spiking Neural Networks
The brain-inspired neural architecture in which artificial neurons communicate through discrete electrical pulses—as biological neurons actually do—rather than the continuous activation values of conventional deep learning, enabling temporal dynamics and energy efficiency that may be prerequisites for genuine self-modeling and social cognition.
Biological neurons do not transmit continuous numbers. They fire. A neuron accumulates input until it crosses a threshold, then discharges a spike—a brief, discrete electrical pulse—and resets. The timing of that spike, relative to other neurons in a network, carries information that continuous-valued networks cannot easily represent. Spiking neural networks (SNNs) replicate this architecture, and in doing so they open computational possibilities that conventional
neural networks structurally foreclose. The temporal dynamics of spiking allow information from different time points to be bound together in ways that continuous-activation networks cannot readily achieve; the discrete, sparse firing patterns enable energy efficiency orders of magnitude greater than the dense matrix multiplications of
large language models; and the architecture's proximity to biological neural computation makes it the natural substrate for modeling the cognitive capacities—self-perception, embodied sensorimotor experience, genuine theory of mind—that Yi Zeng's BrainCog platform pursues. SNNs are not merely a more efficient implementation of conventional