Fig. 8: Architecture diagram of the forecasting model.
From: Predicting progression events in multiple myeloma from routine blood work
Architecture diagram of the forecasting model. The Long Short-Term Memory (LSTM) network summarizes all observations of a patient’s sequential blood work within its internal state. The Conditional Restricted Boltzmann Machine (CRBM) then generates probabilities of possible outcomes, conditioned on the patient history encoded by the LSTM, through a Gibbs-sampling algorithm, yielding a distribution over possible future states for the input variables. Model outputs can be fed back into the LSTM for recurrent generation of future observations.
