Machine training — what does the AI know today?
Machine learning
Today’s artificial intelligence (AI) has far surpassed the hype in block and quantum computing. This is due to the fact that huge computational resources are easily accessible to an ordinary person. Developers now use it at creation of new models of machine learning and retraining of existing models for increase in productivity and results. The easy availability of High Performance Computing (HPC) has led to a sudden increase in demand for IT professionals with machine learning skills.
Machine training — what does the AI know today?
When you mark a face in a Facebook photo, it’s artificial intelligence that works behind the scenes and identifies the faces in the photo. In some applications, facial tagging is now omnipresent and displays images with human faces. Why only human faces? There are several applications that detect objects such as cats, dogs, bottles, cars etc. E. On our roads there are standalone cars that detect objects in real time to drive a car. When you travel, you use Google Directions to study traffic situations in real time and follow the best route offered by Google at the time. This is another implementation of the real-time object detection method.
There are several AI applications that we are using almost today. In fact, each of us uses AI in many parts of our lives, even without our knowledge. Today’s AI can perform extremely complex tasks with great accuracy and speed. Let’s discuss an example of a complex task to understand what is expected of an AI application you are developing today for your clients.
Machine learning — traditional AI
The AI’s journey began in the 1950s, when computing power was negligible compared to today’s. The AI began with predictions made by a machine in the same way that a statistician makes predictions using his calculator. Thus, the initial development of AI was based primarily on statistical methods.
Statistical methods
The development of modern AI applications began with the use of centuries-old traditional statistical methods. You must have used direct interpolation in schools to predict future meaning. There are several other similar statistical methods that have been successfully used in the development of so-called AI programs. We say “so-called” because the AI programs that we have today are much more complex and use methods far superior to those used in earlier AI programs.
Some examples of statistical methods that were used to develop AI applications in those days and are still used in practice are listed here-
Кegression
Сlassification
Clustering
Probability theories
Decision trees
Here we have listed just a few of the basic methods that are sufficient for you to begin studying AI, without intimidating you with the breadth that AI requires. If you develop AI applications based on limited data, you will use these statistical methods.
Today, however, there is an abundance of data. Analyzing the vast amount of data that we have is not very helpful because it has its own limitations. Therefore, more advanced methods such as in-depth training are being developed to solve many complex problems.
Teaching without a teacher
In uncontrolled learning, we do not specify a target variable for the machine, but ask the machine, “What can you tell me about X?”. In particular, we can ask questions such as a huge set of X data: “What are the top five groups we can make out of X? Or “What are the functions most commonly found together in X?”. To get answers to such questions, you can understand that the number of data points that a machine will need to define a strategy is very large. In case of controlled learning, a machine can be trained even with several thousand data points. However, in the case of uncontrolled learning, the number of data points that are acceptable for learning starts at several million. Nowadays, the data is usually abundant. Ideally, the data require tutoring. However, the amount of data that flows continuously on a social network is in most cases impossible to curate.
