Design And Implementation Of Bionic Brain Model Building System Based On Brain-Like System

Human beings have never stopped exploring the structure and function of the brain.Brain science and brain cognition are regarded by many countries as the "jewel in the crown" of human science.After the United States,the European Union and Japan,China has launched the brain science research plan in the thirteenth Five-Year Plan national scientific and technological innovation plan,which explicitly proposes the development of brain-like computing and brain-machine intelligence.With the advancement of brain-like computing research,large-scale brain simulations are performed using computers or specialized hardware to simulate the generation of neural signals within neurons,as well as their transmission between neurons.This provides an effective means for humans to study the mechanism of biological brain operation.However,modeling of biological brains to obtain bionic brain models and deploying them to specialized hardware for simulation requires researchers to possess strong computer science expertise,undoubtedly increasing research difficulty for related research.To reduce the knowledge threshold for brain science researchers in modeling and deploying simulations of biological brains,this thesis designs and implements a bionic brain model building system,which provides a platform for relevant researchers to abstract and simulate biological brains.This system mainly includes functions such as model construction,model conversion,model simulation,result display,model management,system management,and log management.These functions provide support for the construction and simulation of bionic brain models and system operation.The main work of this thesis includes:1.Construction of a bionic brain model.This thesis defines the format of bionic brain model related files applicable to the system based on practical application scenarios.This system constructs a bionic brain model based on model parameters such as neurons and neuronal clusters provided by users,and uses structured model description files and model data files to record the structural information and synaptic connection data of the bionic brain model.2.Design and implementation of model conversion strategies.This system is based on a certain type of brain system,which provides specialized brain chips and brain simulation engines to support bionic brain model simulation.Specific hardware has limitations on the number of neuronal connections in the bionic brain model,and neurons exceeding the limit will incur additional time costs for synaptic data reading and spike signal transmission during the simulation process.In addition,during the simulation process,the simulation engine needs to compare the synaptic delay of neurons with the current time beat,and only when the two are consistent will spike signals be transmitted between neurons.The varying synaptic delays between neurons can increase the judgment branch of the simulation engine,resulting in additional time overhead.This thesis designs two model conversion strategies targeting specific hardware to address the above issues,and conducts targeted conversion of bionic brain models to fully utilize the performance of simulation hardware and accelerate the model simulation process when using specific hardware for simulation.3.Extension of neuron model library for NEST simulator.This system integrates the popular NEST simulator in the industry as an alternative simulation engine for users.However,the NEST simulator lacks a classic LIF neuron model supported by a certain type of brain system.The simulation results of different simulation engines on the same bionic brain model should be consistent.To ensure system compatibility,this thesis modifies some of the NEST source code and adds a classic LIF neuron model to meet the neuron type requirements of users when building and simulating bionic brain models.The system in this thesis adopts a front-end and back-end separated development architecture.The back-end includes the business server that uses the Django framework to handle user initiated requests on the front-end and several computing nodes that are responsible for model building and simulation tasks.At the same time,the COS object storage service provided by Tencent Cloud is used to solve the file storage problems related to the bionic brain model,and the MPI messaging interface is used to realize the communication between the back-end computing nodes of the system and the parallel computing involved in building models.This thesis first introduces the relevant technologies involved in the system development and related knowledge in the field of brain-like computing,and analyzes the functional requirements such as model construction,model transformation,and model simulation of the bionic brain model building system,as well as the non-functional requirements such as system performance and security.Secondly,this thesis designs and implements model transformation strategies for specific simulation hardware.Thirdly,this thesis designs the overall architecture of the system and divides the specific functional modules,as well as briefly describes the interface between modules and data storage.Fourthly,the realization process of each functional module of the system is introduced in detail with the sequence diagram and class diagram.Fifthly,this thesis designs test cases to test the operation of the functional modules of the system,and analyzes the test results.Finally,this thesis summarizes the entire work content and proposes improvement methods to address the shortcomings in the system implementation process.