Deep learning is an area of machine learning that deals with the optimization (training) of deep neural networks. This page tries to explain what Deep Learning means.
Neural networks are a family of machine learning methods that are inspired by the neurons in our brains and the synapses that connect them. Artificial neural networks also consist of interconnected neurons, which are activated based on all their inputs. They then propagate their activation to the other neurons connected to them weighted by the strength of the connection between them. These neurons are often separated into different layers: an input layer that receives the data, a number of hidden layers, and finally an output layer that produces relevant output (e.g. the probability that the input data belongs to a certain class). See the image below for an example of a neural network.
Training the network (learning) consists of finding the optimal set of weights for the connections between neurons so that the network produces the desired output when given some input. Thanks to the hidden layers, the network can learn complex patterns in the input data.
Deep Neural Networks
Deep neural networks are simply neural networks with multiple hidden layers. The more hidden layers a network has, the “deeper” it is. Each layer in the network learns patterns based on the layer before it. So the deeper into the network you go, the more complex are the patterns that are captured by the neurons.
This image below shows an example of a deep neural network that was trained on visual data. It show the increasing complexity of the patterns that are learned: pixel values -> diagonal lines -> faces/cats.
This ability of deep neural networks to automatically learn complex patterns based on the data makes them applicable to a wide range of problems and all different kinds of data.
The Rise of Deep Learning
Deep neural networks are not a new idea. Their potential when it comes to modeling patterns in data has long been recognized, but the problem was how to train them. The deeper the network, the more difficult and computationally expensive it is to train them. Traditional optimization techniques for neural networks simply don’t work well for deep networks. Due to a lack of better ways to train deep networks and due to limited computational power, deep learning hasn’t been used much in the past.
But in recent years, deep learning has been on the rise, and it’s for a couple of reasons:
- New techniques have been developed to efficiently train deep networks, also making use of unlabeled data
- Increased computational power and the use of GPUs for parallel computation
- The huge amounts of data
Thanks to these changes, we can now use deep learning to solve big real-world problems. In the last couple of years, deep learning approaches have been breaking records on well-known benchmarks in various fields, such speech recognition, natural language processing and object recognition. Big companies like Google, Facebook and Baidu have also recognized the power of deep learning and are spending a lot of time and effort on deep learning to extract useful information from their data and improve their products.
In conclusion, deep learning is a very powerful machine learning approach that can be used to analyse large amounts of data of all kinds, and to solve a wide range of complicated problems.
Can we help you?
We’d love to share our big data knowledge with you, please get in touch! Keen on knowing more about us? Xomnia is a big data analytics firm based in Amsterdam, the Netherlands. We are experts in data science, big data engineering and advanced business intelligence. Our services include: big data consultancy & projects, big data training and big data traineeships.