Machine Learning might have been one of the most used words when talking about Big Data at Facebook, TikTok and other Internet-based companies. No wonder the most desired job for our new generation is "AI Developer". But what exactly can we imagine as "Machine Learning" and how is it being utilized?
Let's say we have data of people borrowing money from our company having their height, age, weight, income and maybe some other dimensions. Given this data you could use a fitting machine learning model to predict, if a new customer is likely to pay back their debts or not in order to minimize you risk. Another example could be in categorizing/ scouting up and coming basketball talents, knowing which customers to target based on transactions and purchases on your website, deciding if an incoming mail is/ contains spam or not and so on. Perhaps you already recognize the potential.
Machine Learning, in the following short ML, is actually a sub-branch of Artificial Intelligence everyone might have heard of to this date. The following image explains the categories very well on a visual plane.
As you can already see, A.I. (Artifical Intelligence) is a very broad term and generalizes the ability of a non human technical device program to copy human-like behaviour. The techniques are specified in the so called ML. These include mathematical modelling including probabilites, statistics and logical decision-making. Going even further we also encounter Deep Learning, an even more special sub-field of ML, which we will get to know in one of in the future following articles in this new topic series.
You might ask yourself what ML is at this point, since i haven't explained it yet. Even though you will receive even better explanations in upcoming articles, you can think of ML as a method/ result to build a model, which is based on math and executed by a computer to help you answering every kind of (well most of, okay let's just say a very very broad field of) questions regarding data sets. Since not every questions and problems are identical, we can divide different tasks and approaches in to the following terms:
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
Supervised Learning concentrates on models which already have a dataset with fixed outcomes and try to learn from these results for future inputs given the assumption that following inputs will generate already known outputs. In contrast to supervised we have unsupervised learning. As the name already spoilers what is being targeted, in the latter case there is no output known so we the computer has to bring in an order to the data set that fits the best (What does fit actually mean? Stay tuned for answers). Reinforcement Learning goes even beyond and learns from failures of past modeling trying to build even better models ON THEIR OWN. As general as i kept it, you will see it later in detail (Details explained, but not tooo specific so reading doesn't get boring or difficult). If you try to visualize the process take a look at the following image.
Altough a big part of the process is gathering, preparing, cleaning and processing data we will skip that part since it isn't necessary to understand ML. (Even if it makes up more than half of the time, you get the point hopefully. If not contact me [email protected]) Our goal is to have a ML model which is able to predict outcomes of unknown variables based on given data, so the training and evaluation part is very (very) important and decides if our model will succeed or not (succeed in relative terms, accuracy can be defined even better).
In the coming articles i will begin with basic models used in ML like K-Nearest-Neighbours, K-Means Clustering, Neural Networks etc etc so follow me in order to not miss any topic! I am aiming to explain everything as basic as possible so everyone can participate in the world of Big Data and Machine Learning. If this is not the case the in future or already now, ask me anything and i will provide you the best support on different topics.
TL;DR: Machine Learning is the process of mathematical modelling in order to improve decision-making based on dynamical probabilities and statistics.