Deep learning is an advanced method of artificial intelligence that uses neural networks with many layers to process data. Thanks to this deep structure, the system can progressively learn ever more complex and abstract properties from large amounts of information. The approach makes it possible to solve complex tasks that were too demanding for older methods, such as accurately recognising objects in images or fluent translation between languages.
Put simply, deep learning works as learning in layers. Imagine teaching a computer to recognise a cat. In the first layer it learns to identify basic shapes such as edges and corners. In the next it joins those shapes into more complex wholes, eyes or ears for instance. In deeper layers still it assembles these parts together until it is able to recognise the whole cat. It is precisely this “depth” that lets it make sense of data in a way somewhat like a human.
Deep learning
Technical definition: Deep learning is a subcategory of machine learning based on artificial neural networks with multiple hidden layers, referred to as deep neural networks (DNNs). Its characteristic feature is the hierarchical learning of data representations. While the first layers of the network extract low-level features from the data (edges in an image, say), each subsequent layer learns more abstract and complex features on the basis of the outputs of the previous layer. Training such networks typically uses the backpropagation algorithm and large volumes of data. The presence of many layers allows the model to capture very complex, non-linear relationships in the data.
Etymology and context: The term “deep” relates to the number (depth) of layers in a neural network – networks with anywhere from three to hundreds of layers are considered deep. Although the theoretical foundations date back to the 1980s, the massive expansion of deep learning came only after 2010, thanks to the availability of large datasets and a marked increase in computing power, particularly through the use of graphics processing units (GPUs). Deep learning is today the dominant approach in fields such as computer vision (object detection, image segmentation), natural language processing (machine translation, text generation) and speech recognition.