Hybrid CNN-ELM implementation of water quality anomaly detection.
-
Though CNN and ELM are related in their underlying architecture which is a Neural Network, both of them differ significantly in concepts.
-
ELMs are Single Hidden Layer Feed Forward Networks , where their first weight matrix are randomly initialised and they learn only at the last layer. Hence, ELMs have an advantage of faster computation and reduced learning time.
-
In CNNs all the layers learn and this greedy approach ensures joint optimisation.But due to gradient-descent adjustment process it is computationally slower.
-
So we exploit the advantages of CNN and ELM to perform better.
-
CNN is used to extract features and ELM pools the features and makes the final decision.