一个名为Fly.exe的项目成功将雄性果蝇的完整中枢神经系统连接图谱模拟到计算机中运行1。该模拟包含165,122个神经元和25,563,197条突触连接1，配备了逼真的身体物理模型和视觉编码器，在单个RTX 3060 GPU上以15毫秒的耦合间隔执行1。
项目演示了12只虚拟果蝇在30毫米圆形竞技场中的集体行为1。这些虚拟果蝇能够识别并接近食物球体，最终11只果蝇停在食物处，1只停在柱子处，对照组的12只果蝇均未停在食物处1。该结果验证了神经回路在体现式认知中的作用1。
运算效率方面，模拟速度为生物实时的0.015倍，在GPU上完成30秒的仿真耗时约33分钟1。项目采用了参数来源标记框架，所有参数均携带类别标记，包括测量、群体先验、拟合、工程支架和不可恢复等1。项目团队明确声明该模拟不涉及社交行为、觅食行为、个体差异或真实飞行动力学1。
Researchers have developed Fly.exe, a computational system that uploads the complete connectome of a male fruit fly's central nervous system into a computer simulation 1. The implementation encompasses 165,122 neurons and 25,563,197 synaptic connections, representing the full neuronal wiring diagram of the organism 1. The simulation pairs this connectome with a realistic body physics model using MuJoCo and a visual encoder, executing on a single RTX 3060 GPU with a 15-millisecond coupling interval 1.
The project demonstrates the behavior of 12 virtual fruit flies placed in a 30-millimeter circular arena 1. During the simulation, 11 of the 12 flies successfully identified and approached a food sphere before stopping, while one stopped near a pillar, compared to zero flies in the control condition 1. The simulation runs at approximately 0.015 times biological real-time speed, requiring about 33 minutes of computation to simulate 30 seconds of fly behavior 1. All parameters are tracked with source category labels—including measurements, population priors, fitted values, engineering scaffolds, and unrecoverable parameters—following a preregistered validation framework 1.
The researchers explicitly note that the current system does not model social behavior, foraging behavior, individual differences, or realistic flight dynamics 1. The work exemplifies embodied cognition principles by showing how neural circuits can produce adaptive behavior when implemented in a simulated body interacting with a virtual environment 1.
评论
还没有评论，欢迎留下第一条。