What problem does it solve? Designing spiking neural network simulations requires correctly composing neuron dynamics, synaptic filters, projections, delays, and training loops, and mistakes in alignment, units, or state initialization silently corrupt results. This Skill guides an agent through the native brainpy.state modeling path so simulations and surrogate-gradient training runs are constructed correctly. ## Core Features & Use Cases - Point-neuron simulation: Compose LIF, Izhikevich, Hodgkin-Huxley, and other neuron models with unit-aware parameters, explicit State initialization, and brainstate.transform time loops. - Projection and synapse composition: Wire AlignPost/AlignPre projections with event-based communication, exponential or receptor-kinetic synapses, COBA/CUBA outputs, delays, and short-term plasticity. - Surrogate-gradient SNN training: Train spiking networks with surrogate gradients, ParamState selection, JIT-compiled train steps, and complete runnable example scripts such as COBA E/I networks and Fashion-MNIST SNN training. - Use Case: Reproduce a Vogels-Abbott balanced E/I network by selecting LIFRef neurons, EventFixedProb communication, and exponential synapses, then simulate 1000 ms of activity with verified shapes. ## Quick Start Ask the agent to build a BrainPy-State spiking network simulation, for example a conductance-based E/I network of LIF neurons simulated for 200 ms with exponential synapses.
