Methodology

The overall structure of the project, with three partially-overlapping stages of experimentation, model development, and applications, echoes the user-centered design process for deep learning. If the practical objective is to develop computational models for trajectories to be deployed in applications, data collection is followed by model development, evaluation and design/evaluation in context, possibly with cycles within and across the stages. Data collection moves from participatory exploration to controlled experimentation. Model development moves through data exploration and summarization, feature extraction, and statistical learning, which eventually helps to choose models and hyperparameters for featureless (deep learning) models. These are subject to the training-validation-test sequence, where the test step is replaced by user studies in proof-of-concept applications. In this way, we can know how the models perform in realistic scenarios, and how users adapt to trajectory models in use.