Simulating the Self-Organization of Winner-Take-All Networks

Anatomical studies have shown that strong recurrence in local connectivity is a common feature of the superficial neocortical layers of cat visual cortex (Binzegger et al. 2004). The Winner-Take-All (WTA) network is a functional circuit that is in accordance with this type of connectivity (Douglas et al. 1989, Douglas et al. 1995, Douglas & Martin 2004), and is a hypothetical model for the canonical microcircuit. It has been shown to be powerful from a computational point of view (Maass 2000), and also been used in a wide range of applications (e.g. Indiveri 2001, Rutishauser & Douglas 2008, Nessler et al. 2010). We investigate, using the Java-based framework Cx3D (Zubler & Douglas 2009), how this type of network can develop and configure itself in a biologically plausible way. Our simulations begin from a single precursor cell, which encodes it's developmental instructions in a 'gene-like' representation (Zubler et al. 2011). Based on processes such as gene regulation, cell proliferation, axonal outgrowth and Hebbian-type synaptic learning, we obtain neuronal connectivity matching experimental observations of pyramidal and basket cells in layer II/III of cat visual cortex, exhibiting electrophysiological features of WTA networks. The received signal is improved by selecting the strongest input and suppressing low-amplitude noise. Since the developmental rules we incorporate are entirely local, we have shown that WTA-like behaviour can self-organize and calibrate without the instructions of an external agent.