What Is Machine Learning and the Problems ML Is Solving

Recently, we've been investing heavily in our Machine Learning, Artificial Intelligence, and Robotics Lab in our skunkworks program at Codelitt Incubator. A lot of people, by this point, have probably heard a bit about machine learning (ML), but don't really know what it is or the problems it can solve -- even less the business value it can provide.

Machine Learning (ML) is a powerful technology that is behind some of the greatest advances in technology over the past decade -- even more so over the past 5 years. The possibilities of applications are endless. While we are still in the infancy stage of a relatively new discipline of software, the innovation and knowledge curve have been steep. Every year results in new tasks that were previously unsolvable; some by computers and some even by humans.

It is,at its simplest form, a tool with which we impart a computer with sensible decision making capabilities. In Machine Learning, as could be gleaned from its name, we are no longer teaching it how to perform a task, but how to LEARN to perform this task.

In Machine Learning, as could be gleaned from its name, we are no longer teaching it how to perform a task, but how to LEARN to perform this task.

These decision making algorithms can take several forms but most follow some very basic data fitting principles. To better explain let’s use an example to show more or less how this works.

A Simple Machine Learning Example

Let’s start with an uber simple scenario where patients are tested for two different indicators for a certain disease, such as tests for certain proteins in the bloodstream and white blood cell count. Each test will have a score ranging from 0 - 10, scaling with severity, also note that even a completely healthy person would test positive for these indicators. Below is a small set of sample data.

Results (positive/Negative) Score 1 Score 2
positive 2.8 7.8
positive 3.3 6.7
negative 0.5 6.0
positive 3.9 7.1
negative 5.0 0.3
negative 3.9 1.4
negative 4.0 3.9
positive 4.9 3.8
positive 3.7 2.9

Sample Data: Here we have the test results of 9 patients, as well as the overall result.

machine learning data table example

Graph: A proper ML algorithm would learn from this trend and get smarter.

While even a trained observer might struggle to make sense of the data above (particularly while looking at thousands of patients and not this overly simplified graph) a properly trained ML algorithm would experience this task much differently. While a person might learn how to understand signs of disease, a ML algorithm would be able to directly learn from thousands of previous cases and make its own conclusions as to what constitutes a sick individual versus a healthy individual; most likely finding patterns that have not been noticed before by humans with our limited data processing power. While doing this task for thousands of patients might be tedious for a person and lead to human errors, a simple machine learning algorithm would greatly outspeed and perform with less errors without breaking a sweat. This is a rather simple example; most problems you will face might contain hundreds of dimensions, not just two, making it impossible for a human to analyze while a ML algorithm could do it well and fast. Machine learning is to "big data," what Javascript was to HTML - it gives it functionality. All the data in the world doesn't mean much if you can't extract insights from it.

Machine learning is to "big data," what Javascript was to HTML - it gives it functionality.

Real-World Machine Learning Use Cases

Machine Learning is being used all around you without you even knowing it. Some fields which benefit from Machine Learning affecting everyday life. A simple one that affects us in the background is ‘spam’ detection. Although spam sometimes gets into our inbox, many times we do not see it (just go check your spam box to see what I mean). Machine Learning has been used to find patterns unique to spam styled emails. Mail servers, such as Gmail, are able to easily filter through emails without keeping track of every new spam email generated. In essence it is looking for select words or types of patterns in order to determine whether the sent mail is spam or not. It also has some drawbacks, such as companies that send many emails with a certain template might end up being flagged as spam by mistake.

Another field where Machine Learning dominates in is finance. Most serious trading agencies in wall street maintain a dedicated machine learning teams working on predictive machine learning algorithms. Here the benefit is staggering. By being able to learn from extremely high dimensional data for past years; a proper algorithm is able to predict future changes in stock values as current events occur. This particular task is used in a continuous form of the algorithm with high powered computers as to always keep it running.

Another field in which you may or may not have guessed Machine Learning to be beneficial in is the field of image recognition. There are many type of image recognition algorithms in use these days. One use of machine learning in image recognition tasks are those that are able to tell the age of a person in a picture. We all age very differently, and even we people have an extremely difficult time telling the age of a person (we don’t recommend trying to test this out with older people you meet, by the way). It is surprising then to see just how well Machine Learning does in this field. By teaching an algorithm about people and their ages the algorithm is able to discern patterns that may not be obvious to most people. I have seen some of these algorithms at work, and it’s remarkable how accurate it can be. Another image recognition field where Machine Learning dominates is in recognizing handwritten digits in a subset field called optical character recognition. This may seem like a simple task but people all have very different handwriting and there are also different ways to write numbers. With Machine Learning we teach an algorithm to recognize characters to the point where it could recognize handwritten characters even when not written legibly. This allows us to have amazing tools such as remote check deposits, something we can all appreciate.

Image recognition is a subset of computer vision, which in turn is a subset of sensory tasks that Machine Learning takes on. The medical field uses the computer vision field of machine learning to detect cancerous tumors and oil companies are using vision combined with drones to detect leaks in pipelines. Voice recognition systems such as Siri and Google now, likewise, are also performed by machine learning. If you want an example of how fast the Machine Learning field is moving, compare old voice recognition attempts by Microsoft on YouTube when it was first released to today’s Soundhoundor Google Now.

The Future of ML

Machine learning is not only an insanely versatile tool but it is also a  tool which is constantly improving as AI technology goes forward. Consistently more and more customers in our skunkworks and R&D program are wanting to experiment with it. Our company has been been investing heavily in our ML/AI and Robotics Lab because of the big, audacious ideas we believe this field will help us tackle.

I hope this brief intro motivates research as to how machine learning could help you in your future projects. If you’re interested in learning more, check out this TED Talk by Jeremy Howard which is a great overview of the progress being made. 

Stay tuned as future posts will go into more detail on machine learning, such as the difference between supervised and unsupervised ML, Artificial Neural Networks and Deep Learning, and computer vision.

Download our Incubator Resources



We’re known for sharing everything!


Save more time, get more done!


Innovate from the inside

Written by
Abel Castilla 18 Nov 2015

Classically trained physicist and programmer with a passion for everything AI (also my favorite movie).


comments powered by Disqus