Spike-Time-Dependent Plasticity and Heterosynaptic Competition Organize Networks to Produce Long Scale-Free Sequences of Neural Activity

Abstract

Sequential neural activity patterns are as ubiquitous as the outputs they drive, which include motor gestures and sequential cognitive processes. Neural sequences are long, compared to the activation durations of participating neurons, and sequence coding is sparse. Numerous studies demonstrate that spike-time-dependent plasticity (STDP), the primary known mechanism for temporal order learning in neurons, cannot organize networks to generate long sequences, raising the question of how such networks are formed. We show that heterosynaptic competition within single neurons, when combined with STDP, organizes networks to generate long unary activity sequences even without sequential training inputs. The network produces a diversity of sequences with a power law length distribution and exponent -1, independent of cellular time constants. We show evidence for a similar distribution of sequence lengths in the recorded premotor song activity of songbirds. These results suggest that neural sequences may be shaped by synaptic constraints and network circuitry rather than cellular time constants.

Networks that learn the precise timing of event sequences

“…Typical computational models of sequence learning employ networks of neurons (Jun and Jin, 2007;Fiete et al, 2010;Brea et al, 2013) or populations (Abbott and Blum, 1996) that are each active for equal amounts of time during replay. However, sensory and motor processes can be governed by networks whose neurons have a fixed stimulus tuning (Xu et al, 2012;Gavornik and Bear, 2014).…”

“…A complementary approach has been used to infer time by fitting a maximum likelihood model to the rates and phases of spiking neurons in hippocampal networks (Itskov et al, 2011). Our approach is most similar to previous studies that utilize discrete populations or neurons to represent serial order (Grossberg and Merrill, 1992;Abbott and Blum, 1996;Fiete et al, 2010;Brea et al, 2013). Namely, we assume that the memory of each individual event duration is learned in parallel with the others as in Fiete et al (2010), in contrast to the serial building of chains demonstrated in the model of Jun and Jin (2007).…”

“…Various neural mechanisms have been proposed for learning the duration of a single event (Buonomano, 2000;Rao and Sejnowski, 2001;Durstewitz, 2003;Reutimann et al, 2004;Karmarkar and Buonomano, 2007;Gavornik et al, 2009), as well as the order of events in a sequence (Amari, 1972;Kleinfeld, 1986;Wang and Arbib, 1990;Abbott and Blum, 1996;Jun and Jin, 2007;Fiete et al, 2010;Brea et al, 2013). However, mechanisms for learning the precise timing of multiple events in a sequence remain largely unexplored.…”

“…Learning in a wide variety of species, neuron types, and parts of the nervous system has been shown to occur through LTP and LTD (Alberini, 2009;Takeuchi et al, 2014;Nabavi et al, 2014). It is therefore natural to ask whether LTP and LTD can play a role in the learning of sequence timing (Karmarkar and Buonomano, 2007;Ivry and Schlerf, 2008;Bueti and Buonomano, 2014), in addition to their proposed role in learning sequence order (Abbott and Blum, 1996;Fiete et al, 2010). …”

Calcium control of triphasic hippocampal STDP

“…Recent computational modelling suggests that the interaction of these processes is critical to establish and maintain appropriate conditions for transient dynamics during cognitive processing, and the examination of a unified model of neural and synaptic plasticity is therefore a critical direction for future theoretical studies [86][87][88][89]. More generally, an examination of the synaptic and neural dynamics generated by the triphasic STDP rule in network models of hippocampal function with more realistic activity patterns, including theta modulation and phase precession, would contribute significantly to the understanding of hippocampal function during putative learning behaviour [90].…”