It was 1959 when Arthur Lee Samuel, American scientist and pioneer of Artificial Intelligence, made the pioneering concept of “Machine Learning” known to the world. After him, the much better-known Tom Michael Mitchell, Director of the Machine Learning Department at Carnegie Mellon University, provided the following, as well as apparently cryptic, definition of “Machine Learning“:
a program is said to learn from experience E with reference to some class of task T and with performance measurement P, if its performance in task T, as measured by P, improves with experience E.
So…what is really behind Mitchell’s words, and how come the concept of Machine Learning is nowadays so much discussed in more and more fields of knowledge?
Let’s try to shed some light together, leaving out the statistical-computational definitions for a moment.
Let’s expand the concept
We could identify Machine Learning with the programming of the Machine (read Computer), in…the absence of programming (aka: actions on the code).
Still cryptic? Let’s try it this way.
Machine Learning aims to educate Computers to learn from experience, that is, to improve their programmatic performance after performing a task or action, even an incorrect one. Amazing how this attitude of Machine, so described, loudly recalls that of…Human.
Yes, because summarily it is precisely this humanization that draws the attention (or, in some cases, the dismay) of great minds in management and academia these days. However, although it may be beyond poetic, the question of the humanizing machine and the challenge of the two worlds is only a small part of what is really happening in the world of Machines, 2018 A.D. Any of us who find ourselves reading about Machine Learning today usually find the concept well embedded and interconnected with others, such as Big Data, Deep Learning, Neural Networks, and…the mother concept: Artificial Intelligence.
Indeed, with all the above parameters it will be worthwhile to come to terms with them, in the enterprise, and as soon as possible! Overwhelmingly drawing the attention of Boards around the world is the increase and improvement in business performance that Machine Learning seems to bring with it – when applied in a winning strategic manner – along with the objective increase in Corporate profits. Of course, this is done according to each individual’s business strategy. As always.
Therefore, since Google has decided to resemble Man more and more – by slavishly studying and analyzing his movements on the Web, but above all by identifying his way of being, personality, needs and preferences more and more extensively – “Machine Learning” really seems to represent the core topic for an ever-increasing number of organizational entities.
The Machine interests, engages, captures. Its new ways of learning, based on the spontaneous reaction resulting from the statistical analysis of data submitted by Man, are now on the lips and on the table of Managers worldwide.
Machine Learning: what is it?
Machine Learning consists primarily of two main modes of execution:
- Supervised Learning. By “supervised learning,” we mean Machine learning based on the submission by Human hands of complete and exhaustive data and exemplifications. Basically, in this application type of Machine Learning, no leeway is left to the Machine, which instead remains in the hands of the Human. In this case, the Machine is given a set of data-categorically related (two) to the inputs but also to the required outputs-precisely for it to return a plausible response that correlates the former and the latter in an analytically acceptable manner.
- Unsupervised Learning. By “unsupervised learning,” we mean Machine learning based on the executive gap deliberately generated by Man. In essence, automatic scope for action is left to the Machine, for which Human supervision is unnecessary. In this case, the Machine is given data inputs that are not divided into any subcategories, leaving it to spontaneously process “hidden patterns” from Man.
These are the two cornerstones of what Machine Learning has stood for since Mitchell’s time, but which today are gradually being modified according to the advances made by the continuous research that the topic has irretrievably attracted.
In fact, we see new application strategies arising, certainly more widely used. Specifically:
- Machine Learning with reinforcement learning. As is the case with Man, this kind of learning (of the algorithm, of the Computer, of the Machine…) is based on the binary system (previously the exclusive preserve of Human) reward/punishment. In analytical terms, the Machine is set to react with a dynamic environment in which it will have to respond to inputs autonomously, aiming to achieve the highest degree of reward, or “self-reward” where it makes mistakes along the way. If she achieves the goal she will be lavishly “rewarded,” while where it does not happen she will suffer real…punishment (aka: penalization). The learning routine is thus based on a passed&failed system, exactly analogous to that of video games – with which you will no doubt all have happened to interact.
- Semi-supervised Learning, or semi-supervised learning. This executive mode of Machine Learning basically consists of a mix between the two modes of supervised and non-supervised learning. The Machine will then be given two categories of data-some provided with corresponding output and some without. The ultimate goal remains the same, namely, the landing of new, previously unknown “patterns.”
As you have all no doubt noticed, some human minds can take flights of fancy when confronted with topics such as Machine Learning and Artificial Intelligence. Needless to deny that, although in the midst of the Digital Era, there is an exceedingly long way to go to learn about the real evolutions of these wonderful means of high application potential. One would almost be inclined to say that we are only at the beginning.
Nevertheless, you may be surprised to learn that you all come in contact with Machine Learning on an almost daily basis.
When you make a query on Google, the return methods that the algorithm uses (aka, the Search Engine Result Page or SERP) to respond to your needs are nothing more than Machine Learning.
Moreover: you will no doubt all have an email account, which you use perhaps for work. Most likely, your email is equipped with a spam filter. Correct? Well, the algorithmic detection of incoming spamming is nothing but Machine Learning. Advanced Machine Learning mode upgrades, on the other hand, can be applied to data protection systems, or anti-fraud systems on the Net (such as fraudulent credit card cloning).
Main application of Supervised Learning mode concerns the bio-medical field and Scientific Research. Here, the algorithm plays a vital role in the prevention of epidemics or in reaction trends related to bacteria., for instance.
It’s not over. Let’s imagine that you all use Netflix, Spotify and Amazon in your daily life. Well, you should know that the system of analyzing and profiling our activities while choosing this or that new series to watch, our favorite music or our product in e-commerce is nothing but (again…) Machine Learning.
Having thus introduced you to the…magical world of Machine Learning, we would now like to not betray our main Corporate mission: learning. We would therefore like to give you, too, the immediate opportunity to try to make… Machine Learning, together with its leading expert. Courage, give it a try yourself. ☺
Try what Google is doing…
In short: the humanization of the Machine continues overbearingly to channel the attention of researchers (and non; have you ever seen the series Westworld?), especially in the Enterprise. Well aware that the only system to dispel the fear of the unknown is knowledge, we decided to learn more about it-especially considering the real and concrete contribution it can potentially make precisely to the world of Digital learning and E-learning.
Follow the research develops along with us. #learnwithus!