Stopping Criteria

To completely support a fully automatic network training some decisions concerning terminating a particular training session are required. In general the training should be stopped when a network generalises best, when the network gives optimal results to data it has never seen before.

cVision incorporates highly sophisticated stopping criteria through partitioning the available data into different sets. The learning data set is used in a familiar way to update the networks brain and the validation data set is used only for decisions. The most important criteria cVision supports are

• Early Stopping
• Generalisation Loss
• Error Fluctuation