Algorithm 1. Greedy search algorithm for Localized stochastic-sensitive autoencoder-radial basis function (LiSSA-RBF) network.

Input:

Number of activity types: p

Number of training samples: N

Training dataset: $(X,Y)$

Output:

The final LiSSA-RBF network

1. Set the number of radial basis function (RBF) layer units r equal to the number of activity types p;

2. Construct and train a LiSSA-RBF network with r RBF layer units;

3. Compute the training mean square error (MSE) of the trained LiSSA-RBF network;

4. If $r<N/2$, let $r=r+1$ and go to step 2, otherwise go to step 5.

5. The LiSSA-RBF network with minimum training MSE is picked as the final network structure.