What comes to your mind first when you hear the word “intelligence”? Albert Einstein? Stephen Hawking? Steve Jobs? Elon Musk? Without a doubt, these people are insanely intelligent and geniuses in their respective fields. However, their contributions to world science will not be able to measure themselves against what the machine learning systems will achieve for mankind in the coming years. It can clearly be said that the next scientific studies and discoveries will be made by machines and not by humans . Let us now leave the past with Einstein and direct our attention to the future. Above all, we try to answer the question of how machines learn?
Machine learning: overview
The term “artificial intelligence” is increasingly used not only in scientific circles, but also in everyday life. Nowadays there is talk of the next big thing that will either take power in the world or bring breakthroughs in all conceivable areas of human life. Grundsächlich is KI generalization, which includes the various processes of learning machines. Machine learning (ML) is the intuitive name for this process.
Basically, machine learning consists of capturing and processing input data, recognizing patterns, and figuring out how the data can be used to make the right decision.
Machine learning: types
- Supervised learning
- Unsupervised learning
- Reinforcing learning
Supervised learning
In supervised learning, an agent is trained using labeled data. The answers are known for the input data, but they are only displayed to the machine after acceptance. The machine is trained to process and issue a decision that should match the expected output as closely as possible. If a machine input is fed multiple times and receives a penalty for getting the wrong answer, it can be pushed in the direction of the correct answer. After the machine has been properly trained, it can give the correct answer at any time to match the expected output.
Let’s look at a real-life example: You go to the library to find a book about AI and you meet a librarian. He is a big tech freak and certainly knows the best publications about AI. It gives you the opportunity to search for books yourself and learn more about supervised learning in real life, and just tells you whether this book is dedicated to AI or not. By following this trial and error method, you will eventually be able to find good literature about AI. This will give you enough knowledge to help other people find good books about AI. In this case you can compare the librarian with labels on the data that always have the correct answers,
Unsupervised learning
This kind of machine learning is super powerful in the real world. It makes it easier for services like Google Photos to automatically classify the people in the pictures into albums. Another example: Facebook and Google send you targeted ads based on the analysis of what people with similar interests or web activities like you buy. You don’t really listen to your conversations or take advantage of your dreams. It’s just AI that can actually be creepier.
In unsupervised learning, you have to force the machines to produce helpful expenses for problems that have yet to be answered. By providing unlabeled data as input, the machines try to find trends and patterns. The process of recognizing patterns and trends means the ability of the learning algorithm to understand the data.
Imagine the following scenario: You bought a huge box of mysterious chocolate during the Halloween sale. Since everything is unwrapped and thrown into the huge box, you make the decision to make chocolate bouquets for some friends. But you have no idea what kind of chocolate you are getting. So you decide to try a whole bunch indiscriminately. In the end, you will be able to recognize a piece of chocolate as soon as you look at it and easily determine which brand it belongs to and whether there are nuts or caramel in it. The chocolate can then be easily divided into categories. And you get the opportunity to pass them on to your discerning friends who only eat certain types of chocolate.
Reinforcing learning
You’ve probably heard of Elon Musk , who developed an AI bot to defeat the best Dota 2 players in the world, or the Google project “DeepMind AlphaGo Zero”. Both were developed based on the reinforcement learning algorithm. It’s crazy to think that artificial systems are not only able to play our games better than we can, but that they actually give us better methods of playing them ourselves. Remember what has already been mentioned: machine learning means receiving and processing an input as well as returning the useful output. If so, where is reinforcement learning used?
Reinforcing learning is based on the basic idea of teaching a system how to minimize the penalty and how to maximize the reward. This is similar to supervised learning with the idea of punishing the machine if the answer is wrong. Basically, the agent learns to get an answer through the trial-and-error approach. The difference is that you don’t know what the correct answer looks like. There is no fixed input / output combination available for reinforcing learning. The algorithm only understands how to achieve a goal, depending on how it is trained and what is set as a reward and punishment.
Can you imagine walking through a blindfolded maze? However, keep in mind that the walls are made of thorns and on the other side is a hospital. You are faced with the task of walking along the path to the other side without knowing the path and can only orientate yourself by walking against a wall or not. First of all, you will encounter a lot of walls and get a lot of injuries. It will be very painful. Very! But every time you’re at the end of the maze, the injuries are handled carefully. After many repetitions, you will be much better at navigating the maze while avoiding the walls. In the end, you have reached your goal: you just sit in the hospital and enjoy your banana pudding. Because you deserve it! You have successfully learned how to go through a labyrinth unharmed without outside help. Only these walls are bad and hospital pudding is good.
Industries affected by machine learning
All of them. Next!
Seriously. With the amount of data growing exponentially around the world, machine learning seems to be the most effective way of analyzing all of this data today (other than the possible use of quantum computers). Machine learning was developed for the following: Input-> Analysis -> Output . There is really a whole area dedicated to processing and extracting knowledge from data (it is also called data science).
It is possible that the AI will be like electricity in the near future . Either company in any industry will either use them in the next 10 years or fall behind very quickly. It is comparable to not asking companies these days whether they use electricity or not. You just have to do it.
Conclusions
- Machine learning essentially consists of capturing and analyzing input as well as developing a useful conclusion or decision.
- Supervised learning is the use of marked data and training an algorithm to set an expected output on a particular input.
- Unsupervised learning works with problems for which no answer is known yet by clustering and sorting the input data according to certain patterns and trends.
- Reinforcing learning is not really geared towards a traditional edition as such, because it works solely on an incentive system. His goal is to minimize negative penalties and maximize positive reinforcement.
If you have any questions about machine learning, its types and possible applications, you can contact the Tech By Light team by email or Q&A.
Leave a Reply