演題詳細
Poster
大規模シミュレーション
Large Scale Simulation
開催日 | 2014/9/13 |
---|---|
時間 | 14:00 - 15:00 |
会場 | Poster / Exhibition(Event Hall B) |
網膜及び一次視覚野の神経活動を再構築する実時間エミュレータ
A real-time emulation system for reproducing neural activities in the retina and the primary visual cortex
- P3-372
- 奥野 弘嗣 / Hirotsugu Okuno:1 川節 拓実 / Takumi Kawasetsu:1 石田 椋也 / Ryoya Ishida:1 八木 哲也Tetsuya Yagi
- 1:大阪大学 / Graduate School of Engineering, Osaka University
To understand the functional roles of visual neurons in the retina and the visual cortex, responses of these neurons under natural visual environments should be investigated. In this study, we developed an emulation platform for reproducing neural activities in the retina and the visual cortex with the following features: real-time reproduction of both graded potentials and spikes with physiologically feasible spatio-temporal properties, and configurable model structure. To achieve above two features, we employed built-in analog circuits, configurable digital circuits, and scalable multi-core processors in their appropriate roles.
The system was composed of a silicon retina with analog resistive networks, field-programmable gate arrays (FPGA), and SpiNNaker chips, which is a multi-core processor designed by the University of Manchester to simulate massive spiking neuronal networks.
The silicon retina and an FPGA simulate the retinal circuits. Built-in analog resistive networks were employed to mimic the syncytial structure of neurons in the outer plexiform layer (OPL). Built-in circuits are efficient for executing predetermined spatial processing. Spatial properties of the inner plexiform layer (IPL) were simulated by digital circuits in the FPGA. Because a large part of the IPL connections is not well understood, configurability of the FPGA is important for reconstructing the IPL circuit. Temporal properties of both the OPL and IPL were also simulated by the FPGA. To reproduce the action potentials in ganglion cells, the Izhikevich model was implemented in the FPGA. The system developed here reproduces responses of major types of retinal neurons at 200 Hz from the image received by the silicon retina. Furtehrmore, interpolation of the graded potential enables one to generate spikes with one ms accuracy.
SpiNNaker chips receive simulated retinal spikes, and simulate spiking neural networks in the visual cortex in real time. Although four SpiNNaker chips were used in this study, additional SpiNNaker chips offer a larger scale simulation.
We demonstrate that the emulation system can dynamically visualize and predict the complex nature of neural images in early visual circuits induced in the natural visual environment.