An inference engine with human memory simulation on individuals´ activities

One of the challenging tasks in AI is to read and do reasoning on languages in an automatic way as the human does. We present an intelligent inference system to predict people´s activities from English sentences in a simulated human memory model. Our system analyses input sentences with refined natural language processing technologies and stores the extracted information as new memory in a simulated human memory model. The new memory is then processed together with old information in self-generated Bayesian networks for inferring new and unified predictions. To ensure that the reasoning is reasonable, we proposed efficient mechanisms to propagate information, to construct and connect Bayesian networks. We use a series of experiment to compare our system to other reasoning systems such as Direct Memory Access (DMAP) system. The results show an improvement of our system on the deduction of more useful information than the compared systems. With a better disambiguation of confusing data in the knowledgebase to avoid some faulty reasoning, our system is also adaptive to different scenarios.