Incremental ELMVIS for unsupervised learning

The ELMVIS method [5] is an interesting Machine Learning method that optimize a cost function by changing assignment between two sets of samples, or by changing the order of samples in one set which is the same. The cost function is learned by an Extreme Learning Machine (ELM) [13, 12, 10], a fast method for training feed-forward neural networks with convenient mathematical properties [11, 14]. Such optimization problem is found in various applications like open-loop Traveling Salesman problem [7] or clustering [4] (mapping between samples and clusters), but not in Neural Networks. ELMVIS is unique in a sense that it combines the optimal assignment task with neural network optimization problem; the latter is optimized at each step of ELMVIS. A recent advance in ELMVIS method [2] set its runtime speed comparable or faster than other state-of-the-art methods in visualization application.