This reposistory contains the code for the paper Psychophysiological Arousal in Young Children Who Stutter: An Interpretable AI Approach, IMWUT 2022
The MI-MIL approach takes the modality-specific bag representations (π΅π ={π₯1π,π₯2π,...π₯ππ}, π = 19,π = EDA, HR, RSP-amp, RSP-rate) of a 20s physiological sensing data as input. As shown in figure below, MI-MIL has four components: (1) modality specific embedding block, (2) modality specific self-attention pooling block, (3) modality fusion Block, and (4) classifier Block. While the first two blocks are applied to each modality π independently, the latter two combine the cross-modality information to generate inference.
Harshit Sharma, Yi Xiao, Victoria Tumanova, and Asif Salekin. 2022. Psychophysiological Arousal in Young Children Who Stutter: An Interpretable AI Approach. Proc. ACM Interact. Mob. Wearable Ubiquitous Technol. 6, 3, Article 137 (September 2022), 32 pages. https://doi.org/10.1145/3550326
Harshit Sharma, SCAI, Arizona State University, hsharm62@asu.edu