Deep learning, also known as deep neural networks, is a machine learning method based on digital representations rather than task-specific algorithms. Deep learning architectures are inspired by the structure of the brain. Improvements in mathematical formulas and increasingly powerful computers have enabled computer scientists to model many layers of virtual neurons. Deep learning software can be trained to recognize patterns in unstructured data, such as digital representations of sounds, images, video and text.
Deep learning algorithms have been applied to computerized vision, speech recognition, natural language processing, audio recognition, social network filtering, machine translation, bioinformatics and drug design. Unlike other machine learning techniques, the performance of deep learning neural networks does not plateau; deep learning performance continues to increase as the system grows and receives more data input. Because deep learning excels at identifying patterns in unstructured data, enterprises can use it to unlock the value of data they already have, revealing patterns they can use to create or improve products and services. For example, a deep learning model that combines a customer’s search queries, browsing history, news preferences, and movie and TV show rankings can provide a more accurate purchase suggestion. Personalized recommendations on a shopping website can improve a retailer’s competitive advantage as well as their customer relationship management.
Some deep learning workloads make heavy use of CPUs, while others make heavy use of GPUs. The unpredictable process of training neural networks (some training takes hours while other training may take days) requires rapid on-demand scaling of virtual machine pools. Kubernetes clusters offer a flexible, cost-effective option for training different types of deep learning workloads.