The least you should know about AI

What is AI? what is Machine Learning, how does that relate to deep, supervised and unsupervised learning, what about reinforcement learning? what's up with computer vision and Natural Language processing then? Speech, Robotics, Expert systems, etc. What about image classification, object detection? Where are the relationships? if any. Those are the questions I will try to answer here.

A good starting point into AI is to do understand it at least at a conceptual level, which is good enough to get us into a conversation. Here we go.

What is Artificial Intelligence (AI)? AI refers to computing systems able to perform tasks normally requiring human intelligence, like visual perception, speech recognition, translation, decision making, etc.

What is machine learning (ML)? ML is a field of AI that gives systems the ability to learn and improve from experience, without being explicitly programmed. A simplified way to see it is: the systems learn to learn, rather than learning to perform a specific task.

What is deep learning(DL)? DL is a ML method or technique based on Deep Neural Networks (DNN). DL methods allow the system to learn from a training set (data representations like pictures, audio or video) and then apply the learning to a new data set. DL can be supervised or unsupervised.

Is DL the only ML method out there? No, there are others, like Reinforcement Learning

What is Reinforcement Learning (RL)? RL is?dynamically learning based on rewards. The system takes a given action and receives a reward for that action. Then it goes back and correct itself in order to maximize the reward. Like your dog :). RL is a bit different than DL in that RL does not necessarily work with Deep Neural Networks (DNN), but when it does, it is called "Deep Reinforcement Learning".

What are Natural Language Processing (NLP) and Computer Vision (CV) then? NLP and CV are popular uses of ML capabilities. For instance, Image classification, which is used in CV, comes from supervised learning. Recommendation systems (like what you get in Netflix, Spotify, or LinkedIn) comes from unsupervised learning. And gaming AI, comes from Reinforcement learning.

What are the other popular uses of ML capabilities? Robotics, Speech, Expert Systems, Planning, scheduling, and optimization, etc.

Here a graphic for you to have in mind.

Let's try to go even deeper. So far a key element here has been the neural networks, which are made up of neurons, so...

What is a Neuron? In simple terms, a neuron is an algorithm, a mathematical operation for which we have an input and an output. A neuron is the basic component of a neural network.

So, if you make the output of one or more neurons, the input of another one, you will have a neural network.

What is a Neural Network (NN)? a NN is an organized set of neurons that aims to recognize relationships in a dataset through a process that mimics how the human brain operates. It has three types of layers: an input layer, a number of hidden layers and an output layer.

What is a Deep Neural Network (DNN)? a DNN is a NN with many hidden layers.

What are Convolutional Neural Networks then? and what about Generative Adversarial Networks, or Recurrent Neural Networks? They are architectures.

What is architecture? The way a neural network is organized is called architecture. Here some examples:

Finally, it all comes down to mathematics. But beyond that, there are many things you can do with neural networks. What is important (at least to me) is to understand what is what. Why? because if you don't it could be all too messy. And I don't like messiness. Now we know that we could use neural networks in a DCN architecture for image recognition as part of the computer vision system I'm working on. If I am working with audio, I might want to use a recurrent neural network architecture instead. Every architecture is especially useful for a given application.

I hope the above helps someone. Have a great week.

David

Header picture: 11 years ago, New York -orobably my first time there. I was invited by Landesk (later acquired by Emerson and even later sold to a private equity firm). They invited me as representative of the Latin American market and I had to wear my country's flag in my jacket's pocket (you can see the white stick coming out of it). For Latin America though, they had only one flag: the Mexican. So, I had the privilege of being Mexican for that whole week :D

Sources: Many, but I would like to very especially highlight this one, one of the best videos I've found on the topic: https://www.youtube.com/watch?list=PLv5t5RJ4K8awFWGB_HfjilutlQu9bPwij&v=oJNHXPs0XDk&feature=emb_title