Skip-thought representations of user responses are more predictive of MCI status
We first conduct MCI predictions based on the original set of transcribed conversational data from Oregon & Health Sciences University (OHSU)19. Details of the data can be found in Materials and Methods section below. Figure 2 illustrates the two main statistical learning approaches compared in this study.
Overview of proposed algorithm for conversational generation and linguistic marker identification using a RL pipeline. Supervised learning pipeline denotes the classical approach by Asgari et al.9. Our approach is summarized in the RL pipeline and involves a feedback loop with the MCI diagnosis agent generating questions to new users for the purposes of predicting their MCI status using a trained ML classifier.
Here, the classical ML approach presented in Asgari et al.9 for identifying MCI…
