An Unsupervised learning based LSTM model: A new architecture

Abstract:- Recurrent Neural Networks (RNNs) have shown good results with real-world temporal contextual data, but for input sequences with long time lags they fail. Long Short Term Memory (LSTM) model was built to address the issue of large time lags in input data successfully. However, LSTM found lacking for the tasks pertaining to lower level cognitive processing, specifically, information processing, storage & recall, and also whether they could learn in an unsupervised manner. Sustained Temporal Order Recurrent (STORE) networks are designed to encode the order of temporal data, and then could recall the encoded data in veridical as well as non-veridical order employing unsupervised learning. In this research we have propose the Fusion of supervised learning base LSTM propose by Jurgen and unsupervised learning based STORE proposed by Grossberg. To alternate between two approaches as well as mimicking brain in information processing during sleep time (internal input) we proposed CCS (Consolidation Control Unit), built on an in-depth cognitive foundation, to overcome the inability of LSTM to learn in unsupervised manner and to work with lower level cognitive processing. We conclude by providing experimental proof of the efficiency of proposed model by comparing it with original LSTM model