A Spiking Neural Network Model of the Lateral Geniculate Nucleus on the SpiNNaker Machine
Sen-Bhattacharya, Basabdatta and Serrano-Gotarredona, Teresa and Balassa, Lorinc and Bhattacharya, Akash and Stokes, Alan B. and Rowley, Andrew and Sugiarto, Indar and Furber, Steve
Abstract
We present a spiking neural network model of the thalamic Lateral Ge niculate Nucleus (LGN) developed on SpiNNaker, which is a state-of-the-art dig ital neuromorphic hardware built with very-low-power ARM processors. The paral lel, event-based data processing in SpiNNaker makes it viable for building mas sively parallel neuro-computational frameworks. The LGN model has 140 neurons representing a `basic building block' for larger modular architectures. The mo tivation of this work is to simulate biologically plausible LGN dynamics on Sp iNNaker. Synaptic layout of the model is consistent with biology. The model re sponse is validated with existing literature reporting entrainment in steady s tate visually evoked potentials (SSVEP) --- brain oscillations corresponding t o periodic visual stimuli recorded via electroencephalography (EEG). Periodic stimulus to the model is provided by: a synthetic spike-train with inter-spike -intervals in the range 10 -- 50 Hz at a resolution of 1 Hz; and spike-train o utput from a state-of-the-art electronic retina subjected to a light emitting diode flashing at 10, 20 and 40 Hz, simulating real-world visual stimulus to t he model. The resolution of simulation is 0.1 ms to ensure solution accuracy f or the underlying differential equations defining Izhikevich’s neuron model. U nder this constraint, 1 s of model simulation time is executed in 10 s real ti me on SpiNNaker; this is because simulations on SpiNNaker work in real time fo r time-steps dt >= 1 ms. The model output shows entrainment with both sets of input and contains harmonic components of the fundamental frequency. However, suppressing the feed-forward inhibition in the circuit produces subharmonics w ithin the gamma band (> 30 Hz) implying a reduced information transmission fid elity. These model predictions agree with recent lumped-parameter computationa l model-based predictions, using conventional computers. Scalability of the fr amework is demonstrated by a multi-node architecture consisting of three `node s', where each node is the `basic building block' LGN model. This 420 neuron model is tested with synthetic periodic stimulus at 10 Hz to all the nodes. The model output is the average of the outputs from all nodes, and conforms to th e above-mentioned predictions of each node. Power consumption for model simula tion on SpiNNaker is << 1 W.