Figure 3
A flowchart representing the online learning architecture of a sigmoid-modulated rate system. The process begins with the input of an observation. The E-Step computes the posterior. The calculated KL divergence, or surprise, determines the adjustment of the learning rate eta, which is constrained to the interval 0.005 to 0.25. The updated statistics are then computed using an Exponentially Weighted Moving Average (EWMA) and written to memory. The M-Step updates the parameters, which are then fed back to the E-Step through the parameter feedback loop. The memory stores sufficient statistics and is read during the update process. The final output is a prediction.SOHMM online learning architecture