API References
Welcome to the spikeDE API Reference. This documentation provides a comprehensive guide to the internal modules that power our continuous-time Spiking Neural Network framework.
Designed with a Continuous Dynamics First philosophy, spikeDE bridges standard integer-order SNNs with advanced Fractional-Order Calculus, enabling infinite memory and complex temporal dependencies without altering your core model logic.
Core Modules
-
spikeDE.neuron
The foundation of continuous spiking dynamics. Defines stateless neuron modules that compute instantaneous derivatives independent of history.
-
spikeDE.odefunc
The engine for Spiking Neural ODEs. Uses PyTorch FX to symbolically trace and transform discrete SNNs into continuous vector field functions.
-
spikeDE.solver
A differentiable numerical engine for Fractional Differential Equations (FDEs). Implements high-order discretization schemes like Grünwald-Letnikov.
-
spikeDE.snn
The high-level wrapper (
SNNWrapper) that unifies the ecosystem. Converts standard PyTorch SNNs into fractional systems with flexible configuration. -
spikeDE.surrogate
Essential tools for training SNNs via backpropagation. Provides smooth approximations for the non-differentiable spiking operation.
-
spikeDE.layer
Specialized output modules for aggregating spatiotemporal spiking activities. Provides layers to prepare task-ready predictions.
Design Philosophy
The spikeDE architecture is built on the separation of concerns:
- Dynamics Definition: Neurons define what changes (\(dv/dt\)).
- State Evolution: Solvers define how it changes over time (integration).
- Graph Transformation: FX traces bridge the gap between discrete PyTorch modules and continuous mathematical systems.
This modular design allows you to upgrade any standard SNN to a Fractional-Order SNN simply by wrapping it, unlocking powerful temporal modeling capabilities with minimal code changes.