Learning about Deep Learning...
Deep Learning (DL), Machine Learning (ML), and even Artificial Intelligence (AI) are hot topics in business and touches every one of us even though it may not always be obvious. My recent interest in learning about DL originates from wanting to update my understanding of the state of the art, reconnect to my high performance computing prior experience, and as a technologist trying to identify its potential application in my own job (e.g. system software engineering). Diving into learning about Deep Learning turns out to be more challenging than I expected and so I thought it would be useful to share my experiences, pitfalls and pointers.
The first challenge of learning about DL is the rather large void between popular media on how DL has solved cool problems and the graduate level content (e.g. Deep Learning by Goodfellow, et al.) and research papers. A Google search of Deep Learning will give a number of great resources (e.g. deeplearning.net, "The Differences Between AI, Machine Learning and Deep Learning" article, etc.) but again most of the content is at either end of the spectrum (eg. popular vs graduate level). Udacity's free online course "Intro to Machine Learning" starts innocent enough quickly requires a strong understanding of Python plus the curriculum for their nanodegree in deep learning puts it at the harder end of the spectrum. NVIDIA Deep Learning Institute's online courses (and workshops) take a more applied approach which is appreciated but there is some assumption of prerequisite knowledge. Oxford, Montreal and Stanford classes are definitely more at the graduate level.
After surveying the resources above (and many more) the fundamental question arises of what exactly should the learning goals be when it comes to understanding DL. The first goal should be to gain a broad understanding of what DL is and how it relates to ML and AI (and other relevant computer science techniques). Second to learn what business problems are most likely to benefit from DL. Third what the workflow looks like to approach such problems. Lastly how to execute on such a workflow to gain familiarity on the application of DL to real problems. These goals can and should be accomplished with different audiences in mind - the hobbyist, the practitioner (e.g. data scientist), the engineer, decision makers, and the researcher. Same goals but fundamentally different content and approaches to learning.
To be continued.... Part 2 - Deep Learning - Getting Started
About the Author - Darrin Johnson has more than 25 years of experience in the computer industry largely in system software engineering with a passion for innovation. His latest quest is to understand Deep Learning and more importantly how it can be leverage to drive innovation. Any comments, suggestions, etc. will be appreciated.