In this series we work towards deep learning. To paint a clear picture of deep learning, however, we first discuss machine learning and neural networks. To explain deep learning clearly, it is important to outline the connections between all these concepts. In this introduction we give a brief overview of the concepts machine learning, neural networks and deep learning, and the connections between them.
Artificial Intelligence (A) is intelligence exhibited by machines. The term AI is often used when a machine mimics cognitive functions that are associated with humans. But where does this intelligence come from? How come AI systems can play chess, hold a conversation and drive a car? It is difficult or even impossible to formulate formal rules with sufficient complexity, that accurately describe our world. It is therefore necessary to give AI systems the opportunity to acquire their own knowledge by extracting patterns from data. This skill is known as machine learning.
Text continues below the image.
Machine learning refers to any system where the performance of a machine in performing a task improves by gaining more experience in performing that task. By feeding an ML algorithm with data (examples), the algorithm gets better and better at its task. In other words, by gaining experience, the algorithm learns. We also call this training. Machine learning consists of algorithms that learn thanks to data. In machine learning, a trend line is usually calculated on the basis of data. The system can recognize a pattern in a cloud with points. The system can also use that pattern to make predictions.
In machine learning the system is trained with examples. This training is intended to teach the algorithm to give the correct answers. This is often done through supervised learning. Here the system is not only fed with input (the data), but also with labels, or the desired answer. The goal is that the system teaches itself to translate the input into this desired output. By training, the machine makes fewer and fewer mistakes and can eventually arrive at the correct output even with new input.
Suppose you want to teach an ML algorithm to recognize pictures of birds. By showing the system lots of pictures of birds and telling it that these are birds (the labels), the system learns to judge exactly what a bird looks like. The more photos of birds you feed the algorithm, the better the system will eventually recognize birds, even in new photos that the algorithm does not yet know. This same principle (learning by experience) is also used in other applications. Take an algorithm that has to learn to recognize fraudulent credit card transactions. You feed that to historical data of proven fraudulent credit card transactions. Eventually, the system can make predictions, such as: this is a fraudulent credit card transaction with 97 percent certainty.
Machine learning is, in short, the design of machines that can learn from data. There are different ways to design these machines. Neural networks are an important approach to machine learning. In this case, the algorithms in machine learning are implemented by using the structure of neural networks. These neural networks model the data using artificial neurons. Neural networks thus mimic the functioning of the brain, which forms a biological neural network.
Within ML you find the subcategory deep learning, where algorithms are structured to create an artificial neural network. The beauty of deep learning is that the algorithms can adjust themselves by adjusting their weights. Think back to the example about fraudulent credit card transactions. With deep learning, each layer of the neural network can recognize a specific characteristic, such as high amounts, banks abroad, etc. Does the system come up with an incorrect prediction? Then the network adjusts the weight given to certain neurons. For example, a high amount of money may ultimately prove to be less important for a correct prediction. The next prediction will be more correct in this way.
In summary, machine learning is the field that deals with the design of machines that can learn from data. One of the models that can be used in machine learning is the neural network. Deep learning is a complex form of machine learning in which multiple layers of neural networks are used to design machines that can learn from data.
In our blog you can find more English articles about tech.
You might also like these articles:
What is AI (Artificial Intelligence): An introduction
What is machine learning: An introduction
What is a neural network: An introduction
What is deep learning: An introduction
