Kak neural network

The Kak neural network, which was first proposed by Subhash Kak, is an instantaneously trained neural network that creates a new hidden neuron for each training sample, achieving instantaneous training for binary data and also for real data if some small additional processing is allowed. These networks, therefore, model short-term biological memory. The training algorithm for binary data creates links to the new hidden node that simply reflects the 0 and 1 values in the training vector. Hence there is no computation involved. This network has been successfully used in a variety of applications in finance, pattern recognition, signal processing, and time-series extrapolation.

References

• S. Kak, New algorithms for training feedforward neural networks. Pattern Recognition Letters 15, 1994, pp.295-298.
• S. Kak, On generalization by neural networks. Information Sciences 111, 1998, pp. 293-302.

External links

• FPGA implementation of Kak neural network
• Optical implementation of Kak network
• The Basic Kak network with complex inputs
• Implementing Kak Neural Networks on a Reconfigurable Computing Platform