15.2 Cs Connection Specifications

The Cs connection type contains a computed change in weight term dwt, a previous change in weight term pdw, and a change in weight term that is used in aggregating weight change statistics over time, dwt_agg. The previous change in weight term enables momentum-based learning to be performed.

The Cs connection specification type contains the following parameters:

float lrate
Controls the rate of learning. It simply multiplies the CHL learning term.
float momentum
The momentum, which includes a weighted average (weighted by the momentum parameter) of prior weight changes in the current weight change term.
float decay
The rate of weight decay, if a weight decay function is being used.
decay_fun
There are two weight decay functions defined by default, but others can be added. These two are Cs_Simple_WtDecay, and Cs_WtElim_WtDecay, which are just like their counterparts in the Bp algorithm, to which you are referred for more details, see section 14.1.2 Bp Connection Specifications.