Abstract

Sequence data can be found everywhere: from base pairs in a DNA molecule to characters in handwritten text. We need models for sequences to predict or forecast future data, to remove noise, and to aid understanding (e.g. by identifying underlying latent variables). Markov models are the simplest such models. They can be used in their own right, or composed to form more complex models. In this tutorial I will give a very short introduction to Markov models for discrete valued data (N-gram models) and real-valued data (auto-regressive models). The goal is to give an intuitive feel for these models, rather than an exhaustive exposition. If time permits, I will show how they can be composed with more complex components, such as neural networks, to perform complex tasks.