An Integrated Hierarchical Temporal Memory Network for Real-Time Continuous Multi-interval Prediction of Data Streams

Abstract

We propose an Integrated Hierarchical Temporal Memory (IHTM) network for real-time continuous multi-interval prediction (RCMIP) based on the hierarchical temporal memory (HTM) theory. The IHTM network is constructed by introducing three kinds of new modules to the original HTM network. One is Zeta1FirstSpecializedQueueNode(ZFSQNode) which is used to cooperate with the original HTM node types for predicting data streams with multi-interval at real-time. The second is ShiftVectorFileSensor module used for inputting data streams to the network continuously. The third is a MultipleOutputEffector module which produces multiple prediction results with different intervals simultaneously. With these three new modules, the IHTM network make sure newly arriving data is processed and RCMIP is provided. Performance evaluation shows that the IHTM is efficient in the memory and time consumption compared with the original HTM network in RCMIP.