Machine Learning: As simple as 1-2-3-4

Our Speakers:

Brian Russell,
Vice President
GeoSoftware

Abstract:

Machine Learning, or ML, can trace its origins back over several hundred years, but the crucial algorithms that led to today’s major breakthroughs were developed in the 1960s. It was not until the start of the second millennium that interest in ML really took off, thanks to the development of the internet and incredibly fast computers with large amounts of data storage.

For anyone new to the field, the recent technical papers on ML can appear overwhelming, thanks to the detailed mathematics and complex figures. In this talk, I will explain the basics of ML using several very simple examples, all revolving around the numbers 1, 2, 3, and 4. These numbers refer to the number of points I will be analyzing using the twin concepts of regression and classification. To paraphrase an old song, one is the loneliest number, so it won’t help us very much. Two is a bit better but is still too simple to explain either of these concepts.  However, three turns out to be the fewest number of points that cannot be solved in a linear way using regression. Therefore, we need to find a nonlinear solution, which is provided by ML. Likewise, four is the fewest number of points that cannot be solved in a linear way using classification, so again we need to find a nonlinear solution using ML. I will also explain these problems in the context of the history of ML development.

Even these simple examples can lead to advanced mathematics, so I will use a minimum of mathematics and try to illustrate the concepts using graphs, which are much more intuitive. I will also illustrate the concepts to two simple geoscience problems and extrapolate to more advanced problems.