Deep learning is a branch of machine learning that uses multi-layered networks of computational nodes called neurons, creating an artificial neural network that mimics the way human brains process information.
Deep learning is a popular technique for tackling tasks such as computer vision, speech recognition, drug design, cancer diagnostics, automated translations, chatbots and self-driving cars.
What is it?
Deep learning is a subset of machine learning that uses multi-layered neural networks, capable of producing better results than previous models or even human expert performance. It can identify patterns in the data and apply those patterns to make predictions.
It is widely used today, and is delivering innovation in areas such Tesla’s self-driving car, Google's translation and image search and game engines such as AlphaGo (which beat an expert human at the game of Go).
Deep learning has grown in interest as access to compute power via the cloud has allowed processing and learning from larger volumes of data. Deep learning can solve more complex tasks than traditional methods of data science and get higher degrees of accuracy.
What’s in for you?
Deep learning is enabling enterprises to offer innovative products and services — from autonomous vehicles to personalized news feeds. It can also improve staff productivity, by automating some tasks — for instance, some customer service queries — enabling staff to take on more value-added work.
Deep learning can reduce operational costs, improve your bottom line, increase your workforce productivity and enable you to make better decisions faster.
What are the trade offs?
Not every data problem is suited to deep learning. In many cases, it will be overkill; in others, better tools are available.
Deep learning also produces results that are difficult to explain. If the data sets you’re using for the computation have errors or worse, bias, the results you get will be flawed.
Many of today’s best deep learning models are trained and sometimes executed on custom hardware, which can add to the operational costs.
Deep learning models typically take longer to train than other machine learning tools and the amount of data needed to train them is higher. Given their size, it’s also costly and time consuming to move models around — say from testing environments to production.