Description

This course is a practical introduction to natural language processing with TensorFlow 2.0. In this tutorial you will go from having zero knowledge to writing an artificial intelligence that can compose Shakespearean prose. No prior experience with deep learning is required, though it is always helpful to have more background information. We’ll use a combination of embedding layers, recurrent neural networks, and fully connected layers to perform the classification. The basic idea behind natural language processing is that we start out with words, i.e. strings of characters, that are almost impossible for the computer to meaningfully parse. We can transform these strings into a vector in a higher dimensional space. Different words will be represented as vectors of different lengths and directions in this space, and this allows us to find relationships between words by finding the component of one vector along another. Don’t worry, the TensorFlow library handles all of this, we just have to have some basic idea of how it works. Since this is a type of supervised learning, we also have labels for our text. This allows the AI to compare the relationships between words to the training labels, and learn which sequences of words represent good and bad movie reviews. This would also work for finding toxic comments, fake product reviews… just about anything for which we need a multiclass classification of text – provided we have enough training data and labels. The last step in complexity is to change the final layer that handles the classification. This allows us to actually output text that the AI thinks is meaningful. What’s really special about this is that the neural network starts out not even knowing that letters are a thing, or that we use spaces and punctuation, to producing something that approximates human level writing.