Spiking neural networks
Notes on membrane dynamics, spike timing, rate coding, and accuracy targets for compact classifiers.
Core workspace
Edgepike Impulse job queue
Get ready for the neuromorphic wave. No neuromorphic processor, no problem. Try your ideas on CPU first.
-
ZIP
Upload dataset
Classification dataset as a zip with one folder per class. Download the first dataset for testing in the Download Center.
-
SNN
Build model
Edgepike Impulse trains and generates the deployment package.
-
C
Download code
Get the generated project zip from your user area.
-
RUN
Run anywhere
Use it on your linux laptop.
Log in to queue projects
Accounts keep project uploads, queue status, and result downloads separated per user.
Knowledge (coming soon)
Neuromorphic material for builders
Concepts for spiking neural networks, temporal coding, event streams, embedded deployment, and dataset preparation.
Event data
Dataset patterns for event cameras, temporal chunks, image-to-spike conversion, and label folders.
Embedded inference
Guidance for generated C sources, quantized weights, headers, and MCU deliverables.
Tutorials
Guides and recipes
Practical tracks for creating datasets, generating C projects, testing locally, and integrating the generated API.
Getting started
From dataset upload to C integration
From a classification dataset to a downloadable C project, with a quick local test path and the C calls needed for your own application.
1. Obtain the generated code
- Upload a classification dataset as a zip file. Use one folder per class, and check the Download center for example datasets.
- Keep image resolution modest for CPU-friendly latency. The current tests use small images such as 28x28 pixels.
- Edgepike Impulse designs the network, trains it, quantizes it, and builds the generated C library.
- When the job is done, download the zip with the generated system files, harness, headers, and library.
Tip: if you want readable preview labels, keep class folder names meaningful or include a labels.txt file with one label per line.
2. Try it locally
After downloading and extracting the zip, run the harness against a dataset with the same class layout.
cd system
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main_harness.py --dataset_path path/to/dataset --labels_path path/to/labels.txtThe --labels_path argument is optional, but it makes predictions easier to read. Then enjoy the first local run.
3. Use the C API
The generated project follows the same call sequence shown in main.c.
#include <edgepike_config.h>
#include <edgepike_cortex.h>
cortex_init();
cortex_sense_from_file("test_input.txt");
cortex_run_inference();
int prediction = cortex_compute_output();In your own project, replace file sensing with the input path you need, then use the returned class index or mapped label.
4. Compile in your own project
The harness compiles the generated sources against the static Edgepike Cortex library. Use the same pattern when integrating the generated code elsewhere.
gcc main.c \
./edgepike_impulse/edgepike_config.c \
-I./edgepike_impulse/ \
-L./edgepike_impulse/ \
-ledgepike_cortex \
-o edgepike_impulse.binKeep edgepike_config.c, edgepike_config.h, edgepike_cortex.h, and libedgepike_cortex.a together in the include/library path you compile against.
Download center
Datasets
Give it a spiky trial.
Dataset
lower_case_emnist.zip
Lowercase subset from emnist.
5.6 MB
Download
Logging
Newsletter
Subscribe for project updates, tutorials, dataset releases, and Edgepike Impulse news.
Newsletter
About us
About the website and project
Edgepike Impulse helps you set upfront before the neuromorphic processors wave arrives.
What it does
Takes your dataset, spikes it, generates an SNN, train it, quantizes it, gives it back ready to deployment.
Who we are
We are just a bunch of students with no money for neuromorphic processors, but with a few arduinos and a bunch of ideas.
Contact
Send a message to the Edgepike Impulse team. We are open to suggestions. 🙂
Projects
Community Edgepike Impulse work
Users can show experiments, demos, datasets, and embedded builds created with Edgepike Impulse.
No projects submitted yet
The first shared build will appear here after a logged-in user submits it.