Research On Brain Machine Integration System Based On Transfer Learning

Brain Machine Integration refers to a new hybrid intelligent system which can com-bine the strengths of Biological Intelligence and Machine Intelligence together via brain machine interfaces, and is regarded as one of the most prominent cutting-edge fields in science and technology for the twenty-first century. In recent years, with the rapid devel-opment of brain science and artificial intelligence, it becomes possible for brain machine integration system to obtain more powerful capability to solve complicated problems, which neither biological nor artificial intelligence system can settle alone. At the same time, with the development of brain machine integration system, it will promote the progress of brain science, artificial intelligence, cognitive science and clinical science both in the fundamen-tal breakthrough innovations as well as in the theory applications. The main research topics for brain machine integration system include the neural rehabilitation and the animal robot.As an important part of the integration system, machine system has the advantages in its huge storage capacity and tremendous computing power. However, compared with the ma-chine system, biological system outperforms in its efficiency and low power consumption.How to combine the advantages of these two systems together thus to obtain a more pow-erful integration system - hybrid intelligence, is the key issue and challenge.To solve the problem, our paper puts forward a new method called brain machine in-tegration based on transfer learning. In this integration system, transfer learning can learn some useful knowledge when solving the source problem,and then apply the learned knowl-edge to solve the target problem which is different but related. On the other hand, brain ma-chine interface can build a direct communication pathway between the biological brain and the external machine. In this way,the brain machine integration system based on transfer learning and brain machine interface can gain more powerful intelligence, which is further discussed in the following three parts.First of all, this paper puts forward a framework and architecture for the brain machine integration system based on transfer learning. The general idea is as follows: First, by the imitation of the transfer learning ability from human and animals,this paper develops some transfer learning algorithms to solve the different but related problems, which are of great significance for the research with only a small number of training data. In this way, it is ex-pected to promote the development of artificial intelligence and machine learning. Besides,in the brain machine integration system, the transfer learning ability can happen between the biological module and biological module, biological module and machine module, ma-chine module and machine module to enhance the intelligence of this integration system. It can also provide some new platforms and methods for the brain science, cognitive science and clinical medicine research.Secondly, based on the framework and architecture of brain machine integration sys-tem, this paper designs two integration systems, and the first one is the ratbot system based on transfer reinforcement learning, where a ratbot is a rat with electrodes implanted in the medial forebrain bundle and sensorimotor cortex of its brain. First, the maze problem is modeled as a basic reinforcement learning problem. According to whether the source agent and the target agent are the same, whether the source maze and the target maze are the same, whether the source task and the target task are the same, some transfer reinforcement learning algorithms are proposed. Then, based on the transfer reinforcement learning al-gorithms, we design a ratbot navigation system to test whether the integration system has more powerful intelligence; Finally,we build a computational model to explain and explore the mechanism underlying this intelligence-enhanced brain machine integration system.Finally,this paper expands the brain machine integration research based on transfer learning from the laboratory study of animal to the clinical study of human, and designs a clinical neurological diagnosis integration system based on transfer extreme learning ma-chine. The idea for this integration system is: First, we model the problem of consciousness diagnosis and regulation as a real-time prediction and drug control system. Then, to solve the problem of limited data in clinical research, this paper designs a transfer extreme learn-ing machine based on features and parameters reuse. Then based on the transfer learning algorithm,this paper designs a brain machine integration system for consciousness diag-nosis and regulation,and evaluate the performance in clinical trials. Finally,based on the results and findings, we discuss the neural mechanism underlying human brain conscious-ness.In summary, this paper focuses on the research about brain machine integration system based on transfer learning. We propose a new framework based on transfer learning and design two integration systems. From the laboratory study of ratbot system based on transfer reinforcement learning, to the clinical study of neurological diagnosis integration system based on transfer extreme learning machine,this paper shows the significance of brain machine integration for the research about intelligent systems and neural rehabilitation.Besides, we explain the feasibility and effectiveness of brain machine integration system from the perspective of computational neuroscience modeling and neural mechanisms.