Network Canvas

The Network Canvas provides a visual drag-and-drop interface for designing spiking neural networks. Add excitatory and inhibitory populations, connect them with projections, configure weights and connectivity, then simulate — all from a browser.

Built on React Flow (Xyflow), the canvas supports node dragging, edge creation via drag-connect, and real-time layout updates.

Quick Start

1. Switch to the Canvas tab
2. Click + Exc to add an excitatory population
3. Click + Inh to add an inhibitory population
4. Drag from one node's handle to another to create a projection
5. Click Simulate to run the network
6. View spike counts, firing rates in the results bar

Populations

Each population is a group of identical neurons:

Property	Default	Description
label	auto	Display name (e.g., "Exc 0")
model	LIFNeuron	Neuron model from the 118 available
count	80 (exc) / 20 (inh)	Number of neurons
neuron_type	excitatory	Excitatory (blue) or inhibitory (red)
position	auto	Canvas x, y coordinates

Excitatory populations are shown with rounded blue borders. Inhibitory populations use square red borders.

Projections

Projections are directed connections between populations:

Property	Default	Description
weight	0.1	Synaptic weight (positive for exc, negated for inh→exc)
delay	1.0 ms	Synaptic delay
probability	0.2	Connection probability (sparse random connectivity)

Create projections by dragging from a source node handle to a target node handle. The weight label appears on the edge.

Simulation

The canvas maps populations and projections to the E-I balanced network simulation backend. For networks with excitatory and inhibitory populations:

• Exc→Exc weight = w_ee
• Exc→Inh weight = w_ie
• Inh→Exc weight = w_ei
• Inh→Inh weight = w_ii

Connection probability from projections is used for sparse random connectivity.

Limits: Maximum 2000 neurons per network (browser performance).

Results appear in the status bar: neuron count, spike count, mean excitatory and inhibitory firing rates.

NIR Export/Import

The canvas supports NIR (Neuromorphic Intermediate Representation) format for interoperability with other SNN frameworks:

• Export NIR: Saves populations as nodes and projections as edges in a JSON file compatible with the NIR specification
• Import NIR: Loads a NIR JSON file and creates populations and projections on the canvas

NIR Format

{
  "format": "nir",
  "version": "0.1",
  "nodes": {
    "pop_a": {"type": "LIF", "count": 80, "neuron_type": "excitatory"},
    "pop_b": {"type": "LIF", "count": 20, "neuron_type": "inhibitory"}
  },
  "edges": [
    {"source": "pop_a", "target": "pop_b", "weight": 0.5, "delay": 1.0}
  ]
}

API Endpoints

Method	Endpoint	Description
GET	/api/graph/models	List available neuron models
POST	/api/graph/population	Create a population node
POST	/api/graph/projection	Create a projection edge
POST	/api/graph/validate	Validate network graph
POST	/api/graph/simulate	Simulate network
POST	/api/graph/export-nir	Export to NIR format
POST	/api/graph/import-nir	Import from NIR format

POST /api/graph/population

{"label": "Exc 0", "model": "LIFNeuron", "count": 80, "neuron_type": "excitatory", "x": 100, "y": 100}

POST /api/graph/simulate

{
  "populations": [...],
  "projections": [...],
  "duration": 200.0,
  "dt": 0.1
}

Returns spike times, neuron indices, firing rates, and a graph summary.

POST /api/graph/validate

Returns {"valid": true, "errors": []} or validation errors: - Missing populations - Dangling projection endpoints - Zero-weight projections - Neuron count exceeding 2000 limit - Probability out of (0, 1] range