Data modeling usually involves an array of algorithms, each with different assumptions about the data, hand tuned sets of inputs and parameters, and a tremendous amount of domain expertise. But can we model data if we don’t know anything about it? What assumptions about the data can we make? And can we use these assumptions as building blocks for more general algorithms? In this talk, we will explore the fundamentals of unsupervised learning with streaming data. We will show how these fundamentals are used in modern machine learning techniques and how they can be implemented generally and applied across types of streaming data. And we will show some applications and libraries leveraging them that you can use today.

You will learn:

• What it means to model streaming data that you know nothing about
• How to find structure and extract information from novel streams of data
• How modern algorithms are using these techniques to solve challenging problems
• What types of applications benefit most from unstructured learning
• Concrete demonstrations of the application to streaming prediction, anomaly detection, and classification tasks