Low frequency oscillations facilitate STDP-based pattern learning and decoding

Abstract

Event Abstract Back to Event Low frequency oscillations facilitate STDP-based pattern learning and decoding Recent experiments have established that information can be encoded in the spike times of some neurons relative to the phase of a background low frequency (1-10 Hz) LFP oscillation [1, 2, 3] - a phenomenon referred to as “phase-of-firing coding” (PoFC) [3]. These firing phase preferences could result from combining an oscillation in the input current with a stimulus-dependent static component that would produce the variations in preferred phase [4, 5, 6, 7]. However, it remains unclear whether this phase preference is only an epiphenomenon or if it really affects neuronal interactions - only then could it have a functional role. Here we demonstrate that PoFC at low frequencies has remarkable effects on downstream learning and decoding when coupled with Spike Timing Dependent Plasticity (STDP). Consider the abstract problem of a single neuron equipped with STDP receiving activation patterns via thousands of afferent excitatory fibers. How might the neuron detect a repeating activation pattern that affects an unknown subset of its afferents, under conditions where the duration of the patterns is unpredictable and where the patterns recur at unpredictable intervals? Two conventional coding options would involve converting the activation patterns into spikes using either a Poisson rate-coding scheme or Leaky-Ingrate-and-Fire neurons that generate variable firing rates. However, learning to detect the repeating patterns with STDP using this sort of coding is difficult or even impossible. In contrast, when the afferent population receives an oscillatory modulation current, and remarkably even when a small fraction of its afferents (~10%) exhibits PoFC, we show that a neuron can rapidly learn to respond selectively every time the repeating activation pattern occurs. The ability of STDP to detect repeating patterns had been noted before [8], but not in an oscillatory mode. Here we show how the partial formatting of the spike times by the oscillation, so that they mainly depend on the current input current, can be efficiently decoded by STDP and recognized after learning in just one oscillation cycle. This suggests a main functional role for the low frequency oscillations that are found almost everywhere in the brain. More generally, our study demonstrates that an oscillation and the STDP mechanism, by the existence of their intrinsic timescales, can together have a functional role.