Explaining machine learning to a grandpa

In this blog, the title might give you an idea of what I’m going to talk about and well you’re right! I’m going to explain in the simplest way what is machine learning about. By saying that, let’s start!

First of all, we need to understand and to define some concepts such as artificial intelligence, algorithms and so on before starting with machine learning since they are key to understanding clearly the entire post.

Artificial Intelligence (A.I)

This concept is amazing as it is the skill of computer science or a machine which basically makes computers to “think” and be “smarter”

To make it clear let me expose an example:

Day-to-day in social media, chatbots are used to recognize phrases in order to show helpful content to customers who have common questions. Chatbots are that amazing that an experience chatting with them seems sometimes like talking to a real person.

Algorithm

So what is an algorithm? Simply an algorithm is a set of finite instructions that are executed. For example, plating a seed is an algorithm since it is, of course, a finite set of instructions. We need to follow some steps to finish the task and those steps are what an algorithm is composed of.

Now that we have an explanation about artificial intelligence and algorithms we can move forward to machine learning.

Machine learning

It is an application of artificial intelligence meaning that it provides systems the ability to learn automatically and improve without being explicitly programmed.

To say it is a familiar term let’s say that it is a stuff tagger that takes your description of something and it tells you the label to which it should correspond.

Data in machine learning

To describe data in machine learning first we need to know what type of data applies to machine learning. Unprocessed fact, text, value, images, and sounds. Data is one of the most important parts of all Data Analytics, Machine Learning, Artificial Intelligence, Deep Learning and so on.

Without it, we would not be capable of training any model and all modern research and automation will be fruitless. Data has the following two stages:

• Information: This stage occurs when all the data has been interpreted and manipulated, so it has already a conclusion reached on the basis of evidence and reasoning which is meant for users.
• Knowledge: This second and last stage, occurs thanks to the combination of learning, information, experiences and has the capacity to gain an accurate and deep understanding of someone or something.

There are three types of learning algorithms that are applied to machine learning.

Types of learning algorithms

Supervised learning

This learning algorithm is easy to explain if we go onto a concept in real life. For example, let’s say we are in school our training or learning is “supervised ” by teachers who are looking at what we do, correcting our mistakes and they, of course, teach us how to do new things. Simply that’s what supervised learning is about.

But let’s see also the technical meaning of it since machines learn in a different way.

By using this type of learning, the algorithm is trained giving it characteristics (questions) and tags (answers) which are both combined by the algorithm that is able to make predictions.

There are two types of supervised learning: regression and classification

Regression

It has, as a result, a specific number. If tags are used to be a numerical value through variables of characteristics, a digit can be obtained as a resulting data.

Classification

In this type, the algorithm has different patterns and has as purpose classify elements in different groups. This type of algorithm is not in the capacity to decide which group a value belongs to or which is the result of an operation. It only gets to relate in order to acquire a result.

Strong learner vs weak learner

A strong learner or apprentice is one who, given the sufficient number of samples for analysis, is capable of making an error lower than any limit imposed. On the other hand, a weak apprentice is one who can only limit its error at a certain minimum value, another way of explaining it could be that the first learner is capable of resolving many tasks and the second one can only focus on one task.

Unsupervised learning

As in supervised learning we previously did, for this type of learning, we are gonna give an example of the same kind (real-life example)

Since we are kids our knowledge start growing for what we see and the experiences we have through our life. First, we start by using our senses, then walking what start to improve from our own work, after walking we get capable of jumping, running, swimming, etc… And that’s what basically unsupervised learning is.

Technically, unsupervised learning is a method of automatic learning in which the algorithm or a model fits itself to observations.

Unlike supervised learning, to the unsupervised one, only characteristics are provided without providing to the algorithm any tag. Its functions are catalog, create similarity and having the ability to group.

Reinforcement learning

For understanding in the clearest way possible this type of learning a real-life example is what we need here (as in the above ones)

Let’s put the simplest and the most common thing that we as humans “reward”, for instance, when we have a dog, it learns different “tricks” that amazed us such as, giving paw, going for the ball, jumping, etc… and by doing this he gets “rewarded” with a cookie or a bone which makes them have a repetitive habit “optimizing” it every single time they do it.

Technically, an algorithm performs the same task in the better way possible in order to achieve the best result by “winning points” raising up the score colloquially speaking.

Artificial neural networks

It is a computational model that consists of a set of units, called it “artificial neurons” which are connected to each other in order to transmit signals. The information which is sent through it, cross the neural network producing output values.

To make it easier to understand, take a look at the following image

Each neuron is connected with others through some links. In these links, the output value of the anterior neuron if multiplied for a weight value that makes automatic learning. Its main objective is to resolve problems in the same way as the human brain. Consisting of a series of layers, all neurons in one layer connect to the neurons in the next layer.

Deep learning

It is a set of algorithms of machine learning that basically tries to model high-level abstractions in data. Simply what deep learning do is to teach computers to do what we naturally do.

Deep learning is closely related to a class of theories of brain development proposed by cognitive neuroscientists in the early 1990s.

Let’s give an example of what we naturally do and what we want “something" does like us.

When we are driving a car our senses are alert all the time and we can recognize all the signalization that is on the road such as traffic lights, stop signals, zebra crossings and so on… Well, what deep learning has to do in this field is that thanks to this amazing technology there are driverless cars that seek to act as we do. Nowadays is a fact that cars can do what we do and they make our lives much easier.

With the previous example you might already have an idea of what deep learning is capable of, but if not just look at the following list:

• Automatic speech recognition: This is the first and the most convincing successful case of deep learning, it involves multi-second intervals in order to well-recognized speech.
• Image classification: Real-time face detection and emotion and gender classification.

• Real-time analysis of the behavior of a crowded area, you can see this youtube video so you can see how it works.
• Lung cancer detection: By using deep learning is possible to detect automatically and classify early-stage lung cancer on CT images.
• Visual art processing: It is a progress that has been made in image recognition, what this project focuses on is in increasing the application of deep learning techniques to various visual art tasks.
• Mobile advertisement: Have you ever noticed that your cellphone is always sending ads that have to do with what you search about. Well, this is thanks to deep learning since an algorithm is created based on your searches and what it basically does is finding the appropriate mobile audience for certain ads. This might be always a challenge, maybe you have seen sometimes some ads that do not have anything to do with you and this have a reason, many data must be assimilated and considered what it is really difficult to do it in 100% precision to the algorithm.

There are a lot of projects that use deep learning, machine learning, and artificial intelligence really amazing, and helpful for humanity.

Thanks for taking your time for reading it, hope you liked it!

Sources:

[1]: Daniel Faggela. (November 21, 2019). What is machine learning?https://emerj.com/ai-glossary-terms/what-is-machine-learning/

[2]: Jason Brownlee. (August 16, 2019). What is deep learning?https://machinelearningmastery.com/what-is-deep-learning/

[3]: Atul. (May 22, 2019). What is Machine Learning? Machine Learning For Beginners https://www.edureka.co/blog/what-is-machine-learning/