Machine Learning — more present than you think
A trend that started to thrive specially after the early 90s, Machine Learning has been increasingly used to tackle practical problems whose solution using conventional approaches (such as the design of explicit algorithms) would be unfeasible (if not impossible).
Inspired in the way humans learn, Machine Learning studies aim at making programs capable of automatically learn/improve from data (nowadays typically available in huge amounts) and make predictions based on them. More specifically, instead of following static program instructions, it consists in identifying complex patterns and making intelligent decisions/predictions based on approaches such as probability theory, statistics, logic, recursive optimization, neural networks.
People not familiar with such topics usually expected applications that seem more “futuristic”, as one could consider for example the use of machine learning for autonomous navigation/object tracking (e.g. Google Self-drive Car).
In this context, some examples of machine learning applications present in the real, daily world can be really surprising. One of the most spread use is in search engines, specially for behavioral advertisement. Links suggested by Google Ads, Amazon suggested products, Netflix and Spotify recommendations, Facebook “People you may know” are some classic examples. Additionally, speller corrections, text/speech recognition (‘Ok, Google!’, ‘Hey, Siri!’, ‘Hey, Cortana!’), spam detection are some other good illustrations.
Going further into some ‘more surprising’ or ‘unexpected’ uses of Machine Learning, one will find Stock Trading. More specifically, decisions as whether to buy, hold or sell a stock can be well supported by machine learning based on parameters such as local and global economic situation, supply and demand rates, current and past prices. In the field of medicine, such principles can be of great value for medical diagnosis, by learning from databases of anonymized patient records.
They have been also vastly employed in the sports, where the availability of statistic data about opponents, tactical strategies, efficacy of training methods etc. increases day by day. A good example is the article present in https://medium.com/@jamesledoux/machine-learning-and-the-nfl-field-goal-dde242b6dd39#.wc670ow7w, where it is explained how the probability of success of a field goal can depend on multiples factors — such as temperature, wind speed, field surface, distance -, so that a predictive model can make a difference on the decision of whether trying to kick a Field Goal or not.
In the scope of some promising applications of machine learning, one that I personally consider interesting is the attempt to predict strokes and seizures. It represents a really challenging problem in the field of medical diagnosis, with researches on it starting decades ago. Seizures can be defined as transient aberrations in the brain’s electrical activity, which can be commonly visualized in the patient EEG signals — sudden signal deflection at the moment of a epilepsy seizure, for example. However, brain electrical activity is composed of a huge amount of overlapping signal from multiple areas/neurons, so that identifying features that could possibly allow the prediction of seizures is far from being a trivial task. Maybe advances in machine learning can soon culminate in a solution for this type of problem. Let’s hope and keep ourselves updated!