AI/ML Workloads

Workloads, in general, can be classified as task-oriented processes and applications that extract a finite amount of computing resources and time to complete the task in order to produce an outcome. AI workloads are no different, but describe applications and processes that are based on the core underlying AI techniques of machine learning and deep learning where data and information drive these processes. AI workloads fall into two main categories:  training and inference.

Model training:  Large amounts of data are fed into machine learning models and letting them learn to identify patterns and make predictions.

Model inference:   Brand-new data are fed into trained machine learning models that draw conclusions without examples of the desired result.

AI workloads are identified by their complexity and the type of data processed.  AI workloads deal with unstructured data like images, generic text, and other non-tagged and structured data. This is where machine learning and deep learning algorithms play a critical role in analyzing unstructured data.

• Operating systems need to offer a secure and consistent administration experience with support for AI/ML libraries, hardware architectures, and accelerators.
• IT and container orchestration systems need to offer simplified deployment, scaling, and management of AI/ML training and serving.
• IT automation systems need to build, provision, and manage applications and infrastructure across platforms.