• A guide to using eCraft2Learn's neural network blocks
• 2

Snap! has the ability to make projects that define, train, and use neural network models. All that is necessary are list and arithmetic operations. But even with hyperblocks and the ability to compile blocks the programs will run several times slower than similar programs that use the Graphics Processor Unit (GPU). But speed isn't the only roadblock to integrating neural networks into your projects. A great many complex blocks need to be defined to support training and inference. If one's goal is to understand the low-level mechanisms underlying neural networks this is all fine.

If, however, your goal is to use neural networks in your projects to do some intelligent things with text, sounds, images, and data, then a better alternative is to use Snap! libraries that rely internally upon TensorFlow.js. TensorFlow.js provides a large variety of functions for creating, training, using, saving and loading models. And the library can accelerate your programs by taking advantage of the GPU on your computer (or smartphone).

The ecraft2Learn Snap! library provides blocks that interface with TensorFlow.js to provide easy-to-use access to large parts of TensorFlow functionality. Furthermore, there are many advantages to implementing neural networks in a browser tab. Privacy is ensured since no data leaves your device. Nothing needs to be installed. No servers are needed. And yet relatively large powerful models can run at a reasonable speed on most devices.

This workshop will give you a hands-on introduction to the eCraft2Learn blocks for add neural networks to your projects. There are no prerequisites for this workshop (since the underlying linear algebra and calculus are hidden inside of (unfortunately black-box) blocks). Any laptop or tablet should be all you need. While nearly everything works in all modern browsers, we recommend Chrome or Edge. We recommend exploring the ecraft2Learn Snap! library before the workshop but it is not necessary.