Swift AI is a high-performance deep learning library written entirely in Swift. We currently offer support for all Apple platforms, with Linux support coming soon.
Swift AI includes a collection of common tools used for artificial intelligence and scientific applications:
- NeuralNet
- A flexible, fully-connected neural network with support for deep learning
- Optimized specifically for Apple hardware, using advanced parallel processing techniques
- Convolutional Neural Network
- Recurrent Neural Network
- Genetic Algorithm Library
- Fast Linear Algebra Library
- Signal Processing Library
We've created some example projects to demonstrate the usage of Swift AI. Each resides in their own repository and can be built with little or no configuration:
- NeuralNet-MNIST
- NeuralNet-Handwriting-iOS
- A demo for handwriting recognition using NeuralNet
- Pre-trained; just download and run
- Built for iOS
Each module now contains its own documentation. We recommend that you read the docs carefully for detailed instructions on using the various components of Swift AI.
The example projects are another great resource for seeing real-world usage of these tools.
Swift AI currently depends on Apple's Accelerate framework for vector/matrix calculations and digital signal processing.
In order to provide support for more platforms, alternative BLAS solutions are being considered.
Contributions to the project are welcome. We simply ask that you strive to maintain consistency with the structure and formatting of existing code.
Collin Hundley is the author and maintainer of Swift AI. Feel free contact him directly via email.
If you have a question about this library or are looking for guidance, we recommend opening an issue so a member of the community can help!
If you're looking for for help with deep learning, computer vision, signal processing or other AI applications, you've come to the right place! Contact Collin for more information about consulting/contracting.
Your donation to Swift AI will help us continue building excellent open-source tools. All contributions are appreciated!
