Novel approach to fast learning: smart neural nets

We propose a novel approach for fast and well behaved learning of fully connected feed forward neural nets. This is achieved not by designing fast learning algorithms, but by designing a neural net that learns smart by regular backpropagation. We introduce a new neuron called a labour neuron. A smart neural net is then constructed using the labour neurons together with conventional neurons. Hidden layers of the smart neural net are formed by using the labour neurons alone. The conventional neurons alone are used to construct the output layer or the decision layer of the smart neural net. We compare the learning capabilities of both the smart neural net and the conventional neural net toward the regular backpropagation learning algorithm. Unlike the conventional neural net, the smart neural net not only learns extremely fast but also behaves well during the learning. In other words, smart neural nets are academically smart efficient learners.