Deep learning is a machine learning method that forgoes the need to learn task-specific algorithms. Instead, it focuses on feature learning. Deep learning has been applied to a variety of fields including image classification, speech recognition, natural language processing, social network filtering, image recognition, and machine translation.
Deep learning models are frequently based on neural networks and can be applied to both supervised learning and unsupervised learning tasks. Deep learning models and architectures including deep neural nets, and recurrent neural networks have produced results that, in some cases, compare favorably to those of human experts.
A deep learning method called long short-term memory (LSTM) has significantly advanced the solution to real-world problems such as machine translation and speech recognition. Google Translate uses LSTM in its translation service.