Self-driving cars, Alexa, medical imaging – gadgets are getting super smart around us with the help of deep learning.
But why does deep learning work? Why does Alexa recognize me and how can a car drive itself? Can we explain it with cats?
Sure we can!
Let’s dive in and understand deep learning in 7 steps!
1. What is deep learning?
Deep learning is a branch of machine learning, where algorithms learn independently from excessive amounts of information. Similarly to people, these algorithms get smarter with experience by gathering and processing more and more data.
2. How does deep learning work?
The magic in deep learning networks is discovering the pattern and structure behind vast amounts of data. The computational model consists of multiple layers, called neural networks, where data is processed.
3. What happens in neural networks?
We have 3 elements in the neural network: the input layer, which is the data we want to analyze. At least 2 hidden layers, or nodes, which complete the computation with the deep learning algorithm. In the output layer, we have the calculated result.
If the outcome is far from the expected, the weight of the connections is recalibrated, and the analysis is run again. This is done until the outcome is as accurate as possible. For example, when configuring for voice-recognition (like Siri), the weight of the data is adjusted until it gives back exactly what the speaker says in writing. Of course, this is a brief version, if you want to understand neural network in depth, Dan Becker can give you some great insights here.
4. What is the deep learning algorithm doing?
In 2 words: correlation and reduction.
The algorithm finds information which is similar to one another while getting rid of information which is irrelevant. The layers combine information they receive about the data from the previous layer, define it as relevant or irrelevant, and send the relevant outcome to the next node. Irrelevant information is discarded, thus the information reduced. If the information is undefined it remains relevant.
Let’s say for example that we have a deep learning algorithm to find cats in pictures. When I insert this cat picture, the deep learning algorithm will analyze it pixel by pixel. Some nodes initially will see long green lines, and define them as long green lines. In the next layer, the node receives many descriptions as long green lines. This node will define this feature as grass and realize this is irrelevant to find cats and discards it.
5. Supervised and Unsupervised learning
Deep learning has 2 main forms: supervised and unsupervised.
In supervised learning, we tell the computer what the information we put in is. We get more human input this way, but the calculation can be more efficient. Supervised learning is used with large amounts of well-defined data, like the weather.
Unsupervised learning, on the other hand, works with unlabeled information. It makes predictions on currently available diverse data and finds a pattern in a seemingly disconnected environment.
6. Deep learning is not AI
While deep learning is a branch of artificial intelligence, AI extends way further. AI is supposed to be the imitation of human consciousness and independent thinking process performed by a computer node. Deep learning cannot think for itself- it can only make decisions based on the data and instructions it was fed.
7. Why is it important
Deep learning is all around us. The trained algorithms recognize our faces when taking a photo, make a predictive analysis of consumer behavior, or detect fraud.
Using deep learning enables us to turn huge amounts of data into insights, and increase intelligence in technology. These insights have been used for major breakthroughs in science, medicine, agriculture, and many other areas. In combination with other novel technologies as IoT or robotics, deep learning can help tackle some of the biggest problems of our world, and improve people’s lives.