Grandfather learns about machine learning!
Recently, I participated in a brief discussion with my grandfather about “Machine learning”, the man hears a lot of trends in the media lately and he feels lost and ashamed of his ignorance about it. Good news for him, I take a good introduction in Holberton school on the subject. Bad news for me, the man is a farmer.
Grandfather, I know you love my son and we are living a great experience taught him animals species, so let’s begin with him to demonstrate the logic behind the human conscious. At first he started (my son) by classifying anything that doesn’t has a human shape; two legs, two hands, head, skin … or doesn’t speaks recognized as animals. the idea behind this little logic comes from a brief observation and collection of data from around.
After that, i fed his little mind by comic books and animals toys which it has been a decent amount of data for him to develop the “shape” criteria for recognizing animals more efficiently, so he become able to distinguish the Elephant by its elongated proboscis and big ears from the bird with its small size and wings. at this phase my little son knew that there is different types and species of animals. and my dear grandfather you may feel by now the power of data in building conscious, so more data you access more knowledge you get.
Supervised learning
As we taught a baby, if we took a machine and fed it with the right data and images of animals like an elephant or a bird in different colour, size and situation labelled by the right answer and description of the image content, in order to strengthen recognition behaviour, we could be successful the machine imitates human classification criteria. This is what we call it “supervised learning”; based on a data set already gathered with predefined functionalities such as “form”.
There is no magic in this and if we take an example by giving a machine an image of a “flying pigeon”. The computer makes its decision by the percentage that if the wings recognized in the photo must belong to a bird and not to an elephant, because seeing a flying elephant is not a common image unless you give it Dumbo posters in the learning phase.
The machine and the child is pretty much the same taking data based on features like shape, colour, sound in order to find the similarity on finding the right much in a stored data behaviour.
Supervised learning is more often used in many fields in nowadays technology:
Face recognition: You may notice this technology when you take a picture this little white square on faces, and lately when you posted in Facebook and began to identify suggestion about people on the picture. all this behaviour is based on supervised learning.
Spam classification: Also the magic features on our mailing service provider which filter spam email based on the malicious example labelled as spam or not spam.
Unsupervised learning
Supervised learning is quit similar to the learning behaviour of human been in the early stage of life when the only logic we have is what we learn it from the previous available set of data -Thanks to the human history- but unsupervised learning is when you need to build your own logic to discover something new, build your own pattern to understand a new life experience.
Unsupervised learning is the way to manage available data, without labelling them. machine needs to analyse this data by classifying it based on recognized feature of similarity. like you search for the common line to make decision. the best way to figure out this techniques by taking an example of Youtube recommendation based on your historical view, your history is the data set given to Youtube. dealing with this data set as supervised learning is useless coming to the particularity of each user favourite videos. so building one of the unsupervised algorithm to process this continuous feed of data is the only way to deal with users variety.
Recommendation techniques: many service provider using the unsupervised learning techniques to deliver a quite efficient recommendation to the users. For example e-commerce platform using your purchase history to limits your choice by avoiding the less probability picked commodity. Youtube and Netflix provide a list of recommendation even based on the trends watched videos on your region.
Reinforcement Learning
This is my favourite learning techniques and will be yours. reinforcement learning is the way how we take lesson from our fails or take motivation from our success. Grandfather you may remember when you taught me chess and I was horrible playing it, and you decide to punish me every time I make a wrong move. you need to know it’s the same as reinforcement learning and you would be a great reinforcement teacher.
Machine using this techniques to build their own experience about wrong and right based on the previous fail and succeed every time the computer made a bad decision and knows about it, it give a bad weight to this choice in order to help it self next time when encounter the same situation. In contrary when the machine made a good decision an award weight gives to this decision which make it the best fit for the next move. So as much as you train your machine as much the data set about Wrong/Right becomes very precise.
Video games: video gaming industry using reinforcement learning to train their bot facing a real gamer. Google included DeepMind the very knowing organization how builds the unbeatable AlphaGo computer Go player. I invite you to watch this documentary:
Industrial simulation: some time go through an experience is quite a suicide especially for industrial strategies or science experience, take a benefit of reinforcement learning algorithm is the good way to simulate the real world constraint and optimize the resources to reach a specific goals I invite you to read about: An industrial problem resolved by AI and simulation.
Grandfather you should now get a decent idea about machine learning, types and usage. but the real magic coming with the implementation of algorithm. there’s a lot of light focused on machine learning as the best choice to solve humanity problems. It’s true if we focus to use in construction purpose not for distraction.
All the credit for those people helping me write this article:
