Simply ML

Training the Network
Testing the Network
Initial Screen
Making a Neural Network

Inspiration

Nowadays, unless you've been living under a rock, you've definitely heard of the fancy computer science buzzwords, including "machine learning", or "neural networks". Unless you have incredible resources or motivation, these words will continue to mystify you until you pour hours into understanding the concepts and mechanics behind the principle of machine learning (Trust us, we've been there and done that). Our goal is to come up with an application that gives users the ability to change a few parameters here and there, giving them a high-level overview of machine learning that is nevertheless quite informative.

What it does

Our application allows you to build your own neural network at a high level, giving users the ability to control the type and number of layers in the network in each run of the program. The application allows you to submit your own data to train and test on, then uses TensorFlow to build several informative graphs on the neural network as it progresses through the stages of forward and back-propagation.

How we built it

We built the framework of the application using ReactJS and the main page of the application using Canvas (vanilla Javascript). The machine learning is done with TensorFlow JS.

Challenges we ran into

We ran into many problems with integration, especially between the vanilla JS with React, and different TensorFlow JS functions.

Accomplishments that we're proud of

Designing the scratch for ML! We needed to integrate many different ideas to get this project to where it is, but we learned a lot of different tools along the way. The drag and drop mechanism is simple to use, and the graphs provide an idea of how ML algorithms train without the burden of having to code them.

What we learned

Our team members came together with different levels of technical experience, but we all came out with a much better grasp and experience with React, TensorFlow, and building full-stack applications.

What's next for ML Outreach

We plan to build more useful and cool visualizations as well as more options in machine learning, such as linear or logistic classifiers, or even unsupervised learning algorithms.