Machine Learning: An Intuitive Definition

Featured article by Angelica St. Rose, Independent Technology Author

‘Machine Learning’ is a buzzword that has been thrown around incessantly in recent years, with a plethora of different implementations across a vast array of industries. Machine Learning algorithms are being used in risk analysis in the insurance sector, in personalized health monitoring in hospitals and doctor’s offices, in predictive market analytics, a service with practical applications prior to selling your home through programs such as a flat fee mls, and even in your favorite online retail outlet, with sites such as Amazon specifically honing the suggested items you see based on previous search history and purchases.

What is it, exactly? Sure, we hear the word a lot, but other than realizing it has something to do with machines (what is a machine, even?) and learning, what does this phenomena describe? To put it in simple, ELI5, fashion, Machine Learning, a subset of Artificial Intelligence, is essentially a computer program designed to learn on its own. Now, you may be wondering why on earth do we need Machine Learning when teaching a human is far easier and more cost-efficient than teaching a computer? My favorite example came from a graduate Machine Learning class I took in college: consider a doctor. Doctors are trained extensively in medical school so that, one day, they can deliver diagnoses that are accurate.  However, there comes a threshold that limits the amount of information a human brain can retain. With a wealth of information that is constantly evolving, keeping up with and internalizing new information in meaningful ways, may prove to be an impossible feat for a human medical professional.

Unsurprisingly, the cost of misdiagnosing patients is a significant one, and can even go so far as to lead in a loss of life (inconsistency in diagnosing an illness is a scarily common flaw in the medical profession right now!). However, give a program trained in reproducing accurate diagnoses the same inputs, and it will always output the same, correct result. Give the program hundreds of thousands of medical journals based on new information, and it will learn it in a fraction of the time a human agent ever could.  Additionally, with a computer, it’ll never suffer from potentially performance-altering grievances such as: annoyance, anger, sleep-deprivation, and more.

However, that’s just one example. Machine Learning is a part of a constantly growing network of interesting research topics. Today we’ll be covering two.

Deep Reinforcement Learning

Deep Reinforcement Learning is a new adaptation within Machine Learning, with the focus being more so on determining a goal and maximizing it. An agent is placed in an environment with no knowledge of the landscape that surrounds it; instead, it must navigate through the environment in a “trial and error” fashion. Just like you would incentivize a child for taking good actions with candy and toys, you would also spank them or put them in time out for being bad-mannered. This is known as a reward system in reinforcement learning, where positive rewards are used to provoke the agent to make good actions, and negative rewards inducing the contrary. After a period of ‘learning’, the agent should then be able to navigate through an environment while taking the most optimal series of actions to ensure a maximal reward at the end.

The subfield of Deep Reinforcement Learning is beginning to see some truly exciting strides, making headlines when AlphaGo, a computer program that plays the board game Go, defeated a world champion, and arguably the strongest Go player in history, in March 2016. During the game, AlphaGo played a series of highly inventive winning moves, some of which were so landbreaking they overturned hundreds of years of wisdom with strategies never seen before, thus further educating people on the most analyzed, and oldest game, in history.

Big Data

Data is all around us, sometimes in ways we can’t comprehend! Ever stop to think why a drive-through line will sometimes move faster when it’s especially backed up? Big data. With state-of-the-art technology, fast food establishments will only display menu items that are quick and easy to prepare, as to keep traffic moving.

How about in restaurants? Nowadays, more and more food joints are placing systems that look like tablets on each table to allow the customer to pick out appetizers, the main entree, drinks, and other menu items, all with the touch of a button! Put together your entire meal within a few minutes without having to wait for service. Further, some even include the option of being able to order ‘a la carte’ style as you eat.

Ever been to a casino enough times to notice certain machines rotated around the floor, some facing front and center to assure it the first thing guests see when they walk in? Why is that? You guessed it. According to Lon O’ Donnell, MGM’s first-ever director of corporate slot analytics, big data works hand-in-hand with deciding which machines are seen first by guests. After all, the goal is to get potential customers to stay. To do that? Put the machines that are statistically the most well-received in an obvious location, as to increase the probability of a positive, first experience.

Big Data, as its name suggests, has to do with deriving meaningful information from extremely large collections of structured and unstructured data.