Characterization Of Ai Model Configurations

With the use of artificial intelligence models in biosciences, researchers have a need to characterize AI models trained for specific tasks to support model reuse and documentation. For AI model reuse, the concepts of datasheets, dataset nutrition labels, and model cards have been introduced in the past [1, 2]. This work is motivated by characterizing AI models based on metrics derived from optimization curves captured during the initial training process. These metrics, and additional derived attributes, are then included as characteristics in the AI model cards.