Already Know What is Deep Learning and Its History?

It has been a long time since the idea of ​​robots that can think like humans has emerged, such as Talos in ancient Greek myth. Talos is described as automaton (a kind of robot) made of bronze that was created to protect Europe. The idea continued to grow and continued until the first computer was invented, engineers and scientists wondered if computers would one day be able to “think”.

This curiosity has given birth to a field of computer science called artificial intelligence (AI).Artificial Intelligence). Artificial intelligence is the study of the theory and development of computer systems to be able to perform tasks that were previously only possible for humans.

Machine Learning

With the development of artificial intelligence technology, a branch of artificial intelligence has emerged that has received a lot of attention from researchers called machine learning. Machine Learning learn theory so that computers are able to “learn” from data, machine learning involving various disciplines such as statistics, computer science, mathematics and even neurology.

One algorithm machine learning what is interesting is the artificial neural network, as the name suggests the artificial neural network is inspired from the workings of the human brain (which is simplified).

Intuitively looking for inspiration to make a machine capable of “thinking” from the workings of the brain is a good step as well as wanting to make a tool that can fly by watching how birds fly.

In one of the artificial neural network models called MLP (multi layer perceptron) is known as layersa

number of neurons clones are grouped into one layers then layers one becomes input for layers another. MLP is actually a model (mathematics consisting of compositions of functions from vector to vector.

This model is usuallytrain using gradient-based optimization algorithms such as gradient descentvarious problems arise when the neural network model has many layersone

of the famous problems is called the vanishing gradient.

This problem arises because of an artificial neural network with multiple layers is actually a function that consists of many composition functions so that when calculating the gradient of the parameters of the function, we must use a chain rule that causes the gradient of the parameter to be small so that the algorithm gradient descent walking slowly.

Deep Learning

In 2006, Geoffrey Hinton introduced one of the variants of artificial neural networks called deep belief netsthe idea totrain This artificial neural network model is totrain two layers

then add one layers on it, then train only layers top and so on. With this strategy we cantrain artificial neural network model with layers more than previous models. This paper is the beginning of the popularity of the term deep learning to distinguish the architecture of artificial neural networks with multiple layers.

After the term deep learning popular, deep learning has not been of great interest to researchers because of artificial neural networks with many layers has a large algorithmic complexity, so it requires a computer with high specifications, and is computationally inefficient at that time.

Until 2009 Andrew ng et al introduced the use of GPUs for deep learning through a paper entitled Large-scale Deep Unsupervised Learning using Graphics Processors. By using the GPU the artificial neural network can run faster than using the CPU.

With the availability of adequate hardware development deep learning started rapidly, and produced products that we can enjoy today such as facial recognition, self-driving carvoice recognition, and more.