It is widely assumed that synaptic plasticity provides the neural basis for long-term memory in the brain (Abraham, 2008, Caporale and Dan, 2008, Martin et al., 2000). However, the precise nature of the underlying representation is still unclear, despite being of great relevance to brain research (Caroni et al., 2012). Izhikevich (2006a) has proposed that Polychronous Neural Groups (PNGs) might provide this representational mechanism. Importantly, this proposal links a high-level concept of representation to a model that is expressed at the level of spiking neurons and an empirically observed learning rule called Spike-timing-dependent Plasticity (STDP).
Polychronous groups are connected groups of neurons that exist in large numbers in brain-like network simulations, and are often related to the Hebbian idea of neuronal assemblies (Hebb, 1949). The neurons in a polychronous group can be fired together in a process called PNG activation. However, the activation of polychronous groups is not synchronous but polychronous (i.e. many-timed), occurring at slightly different times as a wave-like sequential pattern of firing that passes through the group. Izhikevich views such activation events as corresponding to the activation of
specific memory representations.
A system of representation that is based on the activation of polychronous groups is an interesting idea that has not been adequately investigated. I therefore examine this idea in a series of experiments and theoretical discussions. The investigation of this claim has also motivated the development of two new tools. The first is a software tool called Spinula that is the foundation of all simulation experiments in this thesis. The second is a new probabilistic method for the study of PNG activation called Response Fingerprinting (Guise et al., 2013c, 2014). Using these tools I show that PNGs can meet a number of demanding criteria that I deem as necessary for a robust PNG-based representational system.
