Many AI tasks require choosing among predefined options rather than generating an open-ended response. Decision models score those options, offering a way to assess uncertainty and guide automation or human review. When validated for a specific task, these scores can help reduce the risk of acting on unsupported conclusions. Compact decision models can run locally on a CPU, reducing hardware requirements, power consumption, and operating costs.
This session introduces emerging Transformer-based model families, including tabular and time-series foundation models, and examines their characteristics and practical applications. It then focuses on decision models. Through examples in document classification, sensitive information screening, request routing, and model or tool selection, we will explore how these models can support practical decision-making.