Convolution based Variational Bayes for Patient Vital Signs Modeling with Factorial HMM
Abstract: We propose a novel convolution based variational distribution and an EM based learning algorithm to scale factorial HMM to long and complex time series. The number of trainable parameters in our model is independent from the length of the input data. Our model is
also adapted to the use of arbitrarily complex state emission distribution and can therefore be used in combination with patient physiological models. We show the ability of our model to disentangle independent additive processes from synthetic data. Our experiments also confirm that our algorithm is able to fit real world patient data more accurately when several independent Markov chains are used compared to a single Markov chain with a larger state space. Our model could thus offer a scalable, interpretable and versatile alternative to latent space time series models such as standard HMM.
Submission Length: Regular submission (no more than 12 pages of main content)
Assigned Action Editor: ~Mauricio_A_Álvarez1
Submission Number: 243
Loading