Using Convolutional Variational Autoencoders to Predict Post-Trauma Health Outcomes from Actigraphy Data