The Approach:
Large-scale simulations
Large-scale simulations permit inclusion of molecular-level details while replicating system-level behavior, thereby providing insight into how processes at lower levels become integrated to give rise to processes at higher levels. For example, neuroscientists would like to know how various cell types in the visual cortex—and their synaptic connectivity—contribute to the computations it performs, such as recognizing a face.
Progress has been made linking the generation of brain rhythms to cellular mechanisms but the task of linking cognition to cellular mechanisms remains.
|
Running a simulation on a neuromorphic chip You observe all 256 silicon neurons' spike rates (yellow dots) and spike trains (white ticks) in real-time; some neurons are silent (gaps). You select a neuron (white square and red ticks) to see its membrane potential (orange trace) and hear it spike ('pop' sound). You can change the neurons' connectivity and observe the resulting behavior: Here, they inhibit their neighbors and synchrony results—a 40Hz rhythm is evident in the population histogram (beneath spike trains; the other histogram shows individual spike rates). [John Arthur 2006]
|
Researchers think about how a computation could occur and codify their hypothesis by building a model that specifies the ion-channel repertoire of each cell-type (physiology) and the synaptic organization of a network of these cells (anatomy). They simulate the model on a computer, which evaluates mathematical formulae that describe the behavior of ion-channels and synapses. If the model fails to predict the way the actual network responds to a novel situation, that falsifies the hypothesis—providing food for thought!
Neuromorphic chips address computers' inherent limitations as neural simulation platforms: their fundamental component is not a logic gate but a silicon neuron.
In our case, instead of using a computer, we run simulations on neuromorphic chips. You do not need to know how a transistor works to use these chips. You simply download the physiological properties of your cells and the anatomical connectivity of your network to the chip through a USB cable, which also sends back the results in real-time. You can select a single cell in the network and plot its membrane potential, observe the activity of the entire array, or examine any level of complexity in between. You can change the model's parameters and rerun the simulation at the click of a mouse.
