Like the adaptive kernel methods, instead of defining a general unique threshold as seen with several classical AD approaches, the proposed approach allocates a set of thresholds corresponding to the various training samples.
When the partitioning ends, the training samples have been divided into several subsets, each corresponding to one class label.
Another way to account for the variations in the execution of a motion pattern is to have several prototypes for each class, e.g. a subset of the training samples [[31]].
Several neural networks were trained with changing both of the network structure and the number of training samples.
To cope with this issue, existing research suggests several different approaches, such as altering the training sample by up-sampling or down-sampling, i.e., balancing.
Two training samples are used.
Training samples are gained by FEM analysis.
For each object we provide 200 training samples.
Fig. 2 Synthetic training samples.
defining thresholds for training samples.
