Is AI different from ML

What is the difference between artificial intelligence, machine learning and deep learning?

11.11.2019

For example, when Google DeepMind’s AlphaGo defeated the champion in the board game “Go”, Lee Se-Dol, the terms AI, machine learning and deep learning were repeatedly mentioned in the media to describe Google DeepMind. All three technologies helped AlphaGo defeat Lee Se-Dol; yet they differ.

The easiest way to show their relationship to one another is to use a concentric circle. AI is the first and biggest idea, followed by machine learning and, last but not least, deep learning continues the development of AI and machine learning.

 

Artificial intelligence has been part of the human imagination since computer specialists first introduced the term at the Dartmouth Conference in 1956, giving birth to the field of AI. Since then, it has been prophesied that AI is the key to a new and better future for mankind.

AI has nearly exploded in recent years, especially since 2015. This is due to the widespread availability of GPUs. They enable fast, cheap and strong parallel processing. The combination of infinitely large memory and data volumes also plays a major role.

A simple idea from computer specialists could develop into a boom in 2015. This boom has given rise to applications that hundreds of millions of people use every day.

Artificial Intelligence - Machines have human intelligence

Computer programs that played checkers are one of the earliest examples of artificial intelligence. At the Dartmouth Conference in the summer of 1956, the AI ​​pioneers had the idea to construct a complex machine that, with the help of the newly emerging computers, possesses the same intelligence as a human. This concept is called “General AI”: machines that have all of our senses, our reason and think like we do.

What can already be implemented from this today is a concept of “limited AI”. Technologies that are able to solve specific tasks just as well or better than humans. An example of this is face recognition from Facebook or image classification on Pinterest.

The technologies show aspects of human intelligence. But how does it work and where does the intelligence come from? This is where machine learning comes into play.

Machine Learning - An Approach to Achieve Artificial Intelligence

Machine learning is basically a process that uses an algorithm to analyze data, learn from the data and make a statement or prediction about it. Unlike software that was programmed by hand and performed tasks using special instructions, the machine is trained using large amounts of data and algorithms. This enables her to learn how to perform a task.

The idea of ​​machine learning comes from the early AI experts. The algorithm has evolved over the past few years, including decision trees, induced logic programs, clustering, reinforcement learning and Bayesian networks. But none of them achieved the goal of “General AI” and even “limited AI” was difficult to achieve with early machine learning approaches.

It turned out that one of the best uses for machine learning was ComputerVision. Still, a lot of manual programming was required for machine learning to work. Programmers wrote classifiers like Edge Detection Filters. The program could now find out where objects began and where they ended, recognize the shape of an object and recognize the letters STOP. From this, an algorithm was developed that recognized images and learned what a stop sign is. So far a good idea, but on days when it was very foggy or when trees partially covered the sign, the algorithm was rarely able to identify the sign. This was the reason why computer vision and image recognition received little attention until now. Only today do the right learning algorithms make the difference.

Deep Learning - The technique to implement machine learning

Finding cat pictures on YouTube was one of the first successful demonstrations of deep learning.

Artificial Neural Networks is inspired by the biology and processes of our brain. But in contrast to the brain, in which neurons can connect to any neuron at a certain physical distance, the neural network can only do so via separate levels, connections and directions of data propagation.

For example, take a picture and cut it up into a bunch of pieces. These are recorded in the first level of the neural network. In this first level, individual neurons pass the data on to the second level. In the second level, this process is repeated until a final result is produced.

Each neuron assigns a rating to its input, whether this is correct or incorrect depends essentially on the task that is to be carried out. The final score is then determined by the overall ratings. Think back to the stop sign example. Properties of a stop sign image are cut up and checked by different neurons - the octagonal shape, the red color, the distinctive letters, the size. The job of the neurons is to find out whether it is a stop sign or not. The neural network outputs a probability vector which is based on the evaluations. In the example with the stop sign, the system could come to the conclusion that it is 86 percent a stop sign, seven percent a speedometer, five percent an object stuck in the tree, and so on. The network architect then gives the system feedback on what is correct and what is not.

Neural networks have been around since the beginning of AI, the problem was that they were very computationally intensive and therefore practically impossible to implement. A small research group from the University of Toronto, headed by Geoffrey Hinton, is continuing to work on the concept, but it was only with the development of the GPUs that they were able to prove its function.

If you take the example of the stop sign again. The system will often give the wrong answer at the beginning because it needs training. It is likely that the system will have to see hundreds of thousands or millions of images before the evaluation is mature enough that the result is always correct.

Deep learning had its breakthrough with the computer scientist Andrew Ng. He enlarged the neural network, increased the levels and neurons, and then fed a large amount of data into the system to train it. The “deep” in deep learning stands for the large number of levels.

Nowadays, image recognition by trained machines is better than humans in some scenarios. One area of ​​application is, for example, the identification of indicators of cancer in the blood or tumors on MRI images. Google’s AlphaGo also learned the board game by training the neural network and repeatedly playing against itself.

Thanks to deep learning, AI has a prosperous future

Deep learning enables many practical applications of machine learning and extensions to the field of AI. Deep learning handles almost all tasks, so that any type of machine assistance appears possible.

Autonomous vehicles, preventive healthcare, film recommendations Artificial intelligence is the present and the future. With the help of deep learning, AI can become the science fiction that you have only imagined so far.

 

Do you also want to benefit from the use of artificial intelligence, machine learning and deep learning? Contact us! Our experts will be pleased to advise you.