BrainGrid Simulation Results: Growth and STDP

Result files from the BrainGrid neural network simulator. BrainGrid neural simulator was used to produce a network of 10,000 neurons arranged in a 100x100 grid layout. First. a cortical growth network was simulated and then a spike-timing-dependent plasticity (STDP) model was run to refine the network. The output files record information about the network at the end of each epoch. The serialization files store the state of the network at the end of a simulation; these files can be deserialized to recreate the network and run another simulation. The output files for the growth simulation have information about radii of connectivity of each neuron, the firing rate of each neuron, x location of a neuron on the grid, y location of a neuron on the grid, neuron ID, spike history, and spike timing and location for each spike. In addition to the outputs specified for the growth simulator above, the STDP simulation's output file also stores the neuron ID of the source and destination neuron for a synapse, and the weight of a synapse. The serial file stored the source neuron, destination neuron ID, and weight for each synapse. The description of each file is specified below: BG-out-growth-100x100-600e-100s: Output file for growth simulation run using the main BrainGrid repository for 10,000 neurons arranged in al 100x100 layout. The simulation was run for 600 epochs where each epoch was 100 seconds long. BG-serial-growth-100x100-600e-100s: Serial file for growth simulation run using the main BrainGrid repository for 10,000 neurons arranged in al 100x100 layout. The simulation was run for 600 epochs where each epoch was 100 seconds long. SB-out-stdp-100x100-1ep-300s: Output file for STDP simulation run using the Summer of Brain repository for 10,000 neurons arranged in al 100x100 layout. The simulation was run for 1 epoch and the epoch was 300 seconds long.本数据集包含BrainGrid神经网络模拟器（BrainGrid Neural Network Simulator）的结果文件。该模拟器用于构建由10000个神经元组成、排布为100×100网格布局的神经网络。实验流程为先模拟皮层生长网络，随后运行脉冲时序依赖可塑性（Spike-Timing-Dependent Plasticity, STDP）模型对网络进行优化。 输出文件会记录每一轮训练（epoch）结束时的网络状态信息；序列化文件则存储仿真结束时的网络整体状态，可通过反序列化操作重建该网络并开展后续仿真实验。 生长仿真的输出文件包含如下信息：各神经元的连接半径、神经元放电频率、神经元在网格中的x坐标、y坐标、神经元ID、脉冲历史，以及每次脉冲的触发时间与位置。 除上述生长模拟器的输出项外，STDP仿真的输出文件还会存储突触的源神经元ID、目标神经元ID以及突触权重。序列化文件则存储了每个突触的源神经元、目标神经元ID以及突触权重。 各文件的详细描述如下： - BG-out-growth-100x100-600e-100s：基于官方BrainGrid代码库运行的生长仿真输出文件，对应100×100布局的10000个神经元网络，仿真共运行600个训练轮次，每个轮次时长为100秒。 - BG-serial-growth-100x100-600e-100s：基于官方BrainGrid代码库运行的生长仿真序列化文件，对应100×100布局的10000个神经元网络，仿真共运行600个训练轮次，每个轮次时长为100秒。 - SB-out-stdp-100x100-1ep-300s：基于Summer of Brain代码库运行的STDP仿真输出文件，对应100×100布局的10000个神经元网络，仿真仅运行1个训练轮次，该轮次时长为300秒。