A simple model of perceptual decision making built from a single two-dimensional integrator. MT drives a noisy 2D LIP ensemble whose recurrent feedback accumulates evidence in any direction; once the integrator crosses an intercept threshold, an output ensemble spikes. Companion to HBB chapter 8.

A spiking neural model of the basal ganglia performing action selection, implementing the direct and indirect pathways through StrD1, StrD2, STN, GPe, and GPi nuclei. Drop-in subnetwork: feed in a utility vector, read out a one-hot selection signal.

A 3D recurrent ensemble that implements a 2D oscillator whose angular speed is gated by the third dimension. A separate input ensemble drives that speed, demonstrating how one neural population can control the dynamic regime of another.

A spiking neural model of the lamprey locomotion control module from the Neural Engineering Framework. A 3D damped-oscillator central pattern generator (CPG) drives 10D muscle tensions through a basis-function decoding, producing the traveling-wave swimming pattern characteristic of lamprey undulation.

A Legendre Memory Unit subnetwork for Nengo. A recurrent linear layer compresses a sliding window of a scalar input into a low-dimensional Legendre state; downstream readouts can then compute arbitrary linear or nonlinear functions of the windowed input. The headline example learns the windowed RMS via PES on a downstream spiking ensemble.

A spiking neural implementation of the Lorenz "butterfly" chaotic attractor. A single 3D ensemble with recurrent connections implements the differential equations of the canonical Lorenz system, producing the characteristic two-lobe trajectory.

A convolutional spiking network trained on MNIST with NengoDL and deployed on Loihi. Demonstrates training a TensorFlow-defined network inside Nengo, then running the trained spiking network on Loihi neuromorphic hardware.

A spiking Actor-Critic reinforcement learning network from Bartlett et al. (2022). Pairs a pluggable state representation, TD learning rule, and an optional Legendre-Delay-Network memory of past rewards and values to learn policies over continuous state spaces. Vendored from upstream rlnet/.

Spiking-neural estimator of a probability density. Trains a Spatial Semantic Pointer mean encoding mu and a Function Inverse Estimator normalization xi from samples, then drives a spiking ensemble whose firing rate at the SSP-encoded query point approximates the PDF.

A NengoGUI-first model that captures live microphone audio with `sounddevice` and injects it into a spiking ensemble. The microphone always feeds the most recent window of audio into the network, so the display stays live regardless of simulation speed. Cross-platform (macOS, Windows, Linux) — the only OS-specific touch is granting your terminal access to the microphone.

A Semantic Pointer Architecture model that answers questions by binding two attributes (colour and shape) into a single semantic pointer and recovering one attribute given the other as a cue. A visual introduction to circular-convolution binding and unbinding with spa.Cortical.

A Semantic Pointer Architecture question-answering model with basal-ganglia action selection. STATEMENT inputs route the visual pointer into memory; QUESTION inputs gate a motor read-out that unbinds memory with the cue. Shows how spa.BasalGanglia and spa.Thalamus pick the right action.

A Semantic Pointer Architecture question-answering model that holds the bound colour-shape pair in a recurrent memory state, then unbinds with a cue presented after the inputs have gone quiet. Extends the plain SPA question-answering example with a spa.State(feedback=1) working memory.

A Semantic Pointer Architecture model that cycles a state vector through a five-step sequence (A → B → C → D → E → A) under basal-ganglia / thalamus action selection, and continuously projects the state onto a cleanup ensemble that lights up whichever vocabulary element the state currently matches. Companion to the routed-sequencing examples in HBB chapter 7.

A spiking neural path-integration subnetwork that continuously updates a Spatial Semantic Pointer (SSP) representation of position from a velocity signal, using velocity-controlled oscillators with attractor dynamics. Wraps the implementation from Dumont et al. (2023).

A spiking neural SLAM subnetwork that continuously integrates a velocity signal into a Spatial Semantic Pointer estimate of self-position, corrects the estimate on landmark sightings, and learns a semantic associative memory mapping landmark identities to SSP locations. Wraps the SLAMNetwork from Dumont et al. (2023).