Understanding Autonomous Systems video series
AI solutions often look promising in a research lab but fail to deliver real-life scenarios. One of the critical issues with deep neural networks-based AIs is the need for vast human-tagged training datasets. Indeed, most advanced deep neural networks often require millions of individual data points to converge and rich, usable quality level. In addition, once those neural networks are deployed, they mostly behave like black boxes. It makes it very hard for operators and process engineers to monitor their stability and reliability as well as to understand the rationale behind their behaviors. This is often referred to as AIs having low “explainability.” Autonomous Systems offer a novel approach that combines machine teaching, deep reinforcement learning, and advanced simulations to build AI-based control solutions. These systems can not only be trained for real-life scenarios but also leverage operators and other process experts to create more stable and explainable solutions.
This introduction video on autonomous systems, the first of a five-part series, describes how to decide whether your use case is appropriate for Autonomous Systems as well as the key underlying technologies that made them possible.