Welcome to The Neural Network Explorer
Welcome to the neural network explorer! This site includes interactive visualizations of how neural networks learn how to classify data.
Neural Networks have gained popularity in machine learning from their ability to perform at difficult tasks such as image classification and speech-to-text conversion. What makes these tasks difficult is the huge variation in data of the same class. Consider, for example, images of cat; cat pictures have different colors and positions, may be obscured by other objects, there may be one cat or many cats. The cat might also be in an image with a dog - is it a picture of a dog or a picture of a cat? These variations make identifying the contents of an image a difficult task. Even Google gets it wrong, especially with increasing image specificity. For example, a search for "cats playing football" includes an image of a cat hunting:Image is a screen shot of a Gooogle image search for "cats playing football".
Another popular task for neural networks is learning how to forecast serially dependent data. The most common applications of this task are to language and music. In this app we explore how a recurrent neural networks learns from and creates new music.