Introduction to Cindicator (CND) Part 2

Introduction to machine learning, neural networks, and deep learning.

 

Machine learning

Machine learning is a part of artificial intelligence that analyses large sets of data in order to recognize patterns that people are likely to miss and make predictions. With predictions, machines can be better than humans even without artificial intelligence and machine learning simply because the machines can process and analyze much larger sets of data than humans can. This being said, machine learning can truly take the capabilities of computers and the way machines can help humans to the next level. One of the classic examples of a machine learning problem is the ability of the machine to differentiate between images of cats and images of dogs, or between different breeds of a certain animal, be it cats, dogs, or some other animals. The obvious extension of this functionality is face recognition that could, for example, help keep cities safe by recognizing people on the wanted lists using security cameras in public locations and inside of various businesses and organizations.

Another example of machine learning is uploading data of, say, one million examples of combinations of height and weight of various people. With that data and machine learning, a machine could later accurately predict weight by just learning about height and predict height without having the weight figure. Machine learning is capable of learning not just from raw experimental and statistical data, but also from event, research, and historical trends. All of this can help machine learning help people identify elements that are missing and help predict what is coming next.

Arthur Samuel, an American scientist who studies computer algorithms and game theory, came up the term “machine learning” in 1959. Samuel has developed software that played checkers and could also teach itself how to play better. The software has been considered one of the very first applications of artificial intelligence. Samuel believed that teaching computers how to play games and then teaching them how to learn to play the games better could have applications across many industries.

The second wave of interest to the concept of machine learning has occurred in 1990s and has been related to all the data that became available with the invention and growing popularity of the Internet. The Internet did not just make data available, it keeps creating ever-growing amounts of data to this day and will continue to do so in the future. This has led to scientists realizing that instead of teaching machines how to do different separate tasks they could try and teach machines to analyze data the way humans would and then give the machines access to the data that is available.

 

Neural networks

Today, machine learning applications work mostly with neural networks. A neural network is a digital network that sorts and classifies information in the same way the human brain does. Essentially, such a network uses probability to make predictions and then uses the predictions it made in the past to self-assess its algorithms and make better predictions in the future, which is how “learning” works with machines and digital networks.

Using data from such networks, machine learning algorithms can determine with a high probability whether a person who wrote a letter is leaving a happy testimonial or is filing a complaint. They can also analyze music and predict whether the music is going to help someone relax or get someone motivated to workout. Even today there are Internet radio stations such as https://www1.brain.fm/ that allow users to choose music based on the type of activity they prefer to do. Lists of activities typically include relaxation, performing physical work, performing mental work, meditating, or sleeping.

 

Deep learning

Deep learning is the most advanced type of computer learning. Deep learning applies algorithms that perform similarly to human brain to neural networks and historical data. Deep learning creates a connection between a given input and an output. The technology is capable of doing that because it can approximate unknown mathematical functions between any two variables if these variables are related at all, at least by correlation or causation. Deep learning can accomplish task similar to those that the machine learning accomplish, with the difference that the tasks are more complex and/or nuanced. For example, deep learning can be a part of facial recognition software and it can detect both faces and facial expressions. It can also identify various objects in image and video streams. For example, it is critical for systems in autonomous driving vehicles to be able to identify various traffic signs, even when these signs are in a bad physical condition. A sign may have spent a prolonged period of time in the sun and its color may have changed a bit, but the software still needs to be able to identify it because it is still a traffic sign not obeying which may cause an accident or loss of life.

Deep learning can also identify gestures of people in videos, detect voices of different speakers, and recognize various emotions in voices.