Microscopy by Machine Learning

Machine Learning (ML for short) is an emerging field of computer science that has huge implications in the modern world. Defined in 1959 by Arthur Samuel machine learning is a “field of study that gives computers the ability to learn without being explicitly programmed”. Though defined as a field decades ago, machine learning has only recently risen in popularity as computers have become powerful enough to handle it.

In the context of microscopy basic non-machine learning algorithms can be used to identify particles based on their structure and color. To do so the programmer must provide a explicit description of the particle in question, such as width, height, and color information. If more the programmer wants to identify a range of particulates they must also explicitly describe in code the characteristics of the particle. It’s very difficult to explicitly describe a particle accurately based on shape and color, thus it is not uncommon for these basic algorithms to fail when presented with more complex particles.

Machine Learning takes a more human approach. A human identifying particulates under a microscope is trained to look for particular particles. Generally training involves sitting them down at a microscope and looking at fibers whilst a trainer guides them. Conceptually, Machine Learning in microscopy is much the same. You train the program by feeding it 1000s of images containing or not containing the particle of interest. The program then characterizes the particle itself, so that when it is presented with new images it is able to correctly identify the particle it is viewing.

Machine Learning has widespread implications across many fields of science, mathematics, and engineering. The TED video below provides a great introduction to how Machine Learning is being used worldwide.