Bayesian Reasoning and Machine Learning Over the past decade, there has been a considerable growth in interest in artificial intelligence and machine learning. In the broadest sense, these fields aim to “learn something useful” about the environment in which an organism finds itself. How to process the information collected leads to the development of algorithms—how to handle high-dimensional data and deal with uncertainty. In the early stages of research in machine learning and related fields, similar techniques were found in relatively isolated research communities. While not all techniques have a natural description in probability theory, many do, and it is the framework of graphical models (a combination of graphs and probability theory) that enables the understanding and transfer of ideas from statistical physics, statistics, machine learning, and information theory. To this extent, it is now reasonable to expect that machine learning researchers are familiar with the basics of statistical modeling techniques. This book focuses on the probabilistic aspects of information processing and machine learning. Of course, no one is saying that this approach is correct or that it is the only useful approach. In fact, one might argue that it is not necessary because “biological organisms do not use probability theory.” Whether this is the case or not, it is undeniable that graphical models and probabilistic frameworks have helped to generate an explosion of new algorithms and models in the field of machine learning. We should also be clear that the Bayesian perspective is not the only way to describe machine learning and information processing. Bayesian and probabilistic techniques come into their own in areas where uncertainty needs to be taken into account. How the book is organized: One of the aims of the first part of the book is to encourage computer science students to enter the field. A particular difficulty faced by many modern students is the limited formal calculus and linear algebra training, which means that the details of continuous and high-dimensional distributions may put them off. In beginning with probability as a form of reasoning system, we hope to show readers how ideas from logical reasoning and dynamic programming, with which they may be more familiar, have natural parallels in a probabilistic setting. In particular, computer science students are familiar with the concept that algorithms are central. However, it is more common in machine learning to view the model as central, and how to implement it as secondary. From this perspective, understanding how to translate a mathematical model into a piece of computer code is central. The second part introduces the statistical background needed to understand continuous distributions and how to view learning from a probabilistic framework. The third part discusses topics in machine learning. Of course, some readers will be surprised to see their favorite statistical topic listed under machine learning. One of the different perspectives between statistics and machine learning is what kind of systems we ultimately want to build (machines that can do “human/biological information processing tasks”), rather than certain techniques. Therefore, I think this part of the book will be useful for machine learners. Section 4 discusses dynamic models that explicitly consider time. In particular, the Kalman filter is treated as a form of graphical model, which helps emphasize what the model is, rather than as a “filter” as is more traditionally seen in the engineering literature. Section 5 briefly introduces approximate inference techniques, both stochastic (Monte Carlo) and deterministic (variational) techniques.
You Might Like
Recommended ContentMore
Open source project More
Popular Components
Searched by Users
Just Take a LookMore
Trending Downloads
Trending ArticlesMore