Anyone even loosely following technology trends in recent years cannot have possibly missed the ever-increasing references to machine learning and artificial intelligence.
At Somatix, too, when naming our offering’s advantages, my colleagues and I often refer to its advanced machine learning capabilities as a key competitive differentiator. Yet are those following up on our innovations clear on what machine learning is all about? Do our prospects fully appreciate the potential of machine learning capabilities?
In a series of posts starting with this one, I’ll address the general definition of machine learning and AI (Artificial Intelligence), and will examine how these technologies are put to use in healthcare today. I’ll wrap up the series with a number of prime examples of machine learning serving as the crucial foundation for disruptive technology.
So where do we start? It’s important to realize that machine learning and AI are by and large synonymous. Both are essentially intended to teach machines to mimic the human brain’s capacity to think. While this post focuses on machine learning, everything said applies equally to AI.
Machine learning is the ability of a computer system to grow smarter through observation and analysis, and to improve by learning from mistakes and experience. Rather than forcibly being fed knowledge by human programmers, the computer is geared with artificial intelligence algorithms enabling it to learn from examples and previous experiences on its own. These algorithms mimic the way in which the human brain’s neural network functions.
The more data the computer accumulates and the more experience it gains, the smarter it
becomes – with every bit of information processed contributing to continuous autonomous improvement of its computational models. The results can be astonishing – smart computers employing machine learning have already reached intelligence levels similar to those of human babies learning by imitating their parents, or of children taught through continuous repetition.
There are two approaches to machine learning:
- Teaching the machine to employ reasoning.
- Modeling the machine on the human brain, and gearing it with computation and data analysis skills that simulate the way in which human brain neurons work.
Researchers can determine machine learning having been achieved by any experience-based improvement of computer performance levels, or by any instance of a computer program performing a task with increasing efficiency in subsequent runs. The better the results produced by a program – with no human intervention – the stronger the indication of computer learning, or machine learning having taken place.
So, if we have computers capable of learning, what wonders can we realize? Well, in a world in which corporate information systems must cope with massive volumes of data (a trend that can only be expected to grow with the Internet Of Things adding countless new information acquisition channels), machine learning can be an extremely powerful tool. It can help filter out “noise” and extract non-explicit information, discover hidden connections, determine models and trends, and even forecast emerging patterns.
Machine learning-based business intelligence, for example, can enable a supermarket chain to analyze the purchasing patterns of a specific consumer segment (a certain age group, for instance) in a given geographical region, as a means of measuring product profitability and campaign efficiency. A healthcare provider chain can similarly leverage machine learning to reveal the cause behind hospital readmissions, by age, gender and geography, among other criteria. In short, machine learning can help organizations uncover previously unknown patterns, and gain insights that contribute to process and profitability optimization.
How exactly does machine learning do all this? The answer lies in its underlying algorithms – sets of predefined calculations dedicated to task-specific forms of pattern analysis. We can divide learning algorithms into two key groups – supervised and unsupervised, both of which aim to solve the same problems. In the case of supervised learning algorithms, the computer is provided with data tagged to represent specific events (enabling it to determine the occurrence of smoking, drinking or eating, for example). In unsupervised learning, on the other hand, the computer is merely given raw data, and is left to comprehend on its own that it represents a mix of distinctly different events (even if it can’t necessarily know which data subset represents which specific event).
Machine learning is a multi-phase rather than single-point process. It encompasses everything from algorithm definition and relevant information detection, through “noise” filtering and knowledge generation, and all the way to output assessment and insight implementation.
Stay tuned for more on AI and machine learning in the next post…
Posted by: Uri Schatzberg, Somatix CTO & co-founder