Spike Timing Dependent Plasticity Finds the Start of Repeating Patterns in Continuous Spike Trains

Timothée Masquelier, Rudy Guyonneau, Simon J. ThorpePLoS 2008 · DOI

Abstract

Experimental studies have observed Long Term synaptic Potentiation (LTP) when a presynaptic neuron fires shortly before a postsynaptic neuron, and Long Term Depression (LTD) when the presynaptic neuron fires shortly after, a phenomenon known as Spike Timing Dependent Plasticity (STDP). When a neuron is presented successively with discrete volleys of input spikes STDP has been shown to learn 'early spike patterns', that is to concentrate synaptic weights on afferents that consistently fire early, with the result that the postsynaptic spike latency decreases, until it reaches a minimal and stable value. Here, we show that these results still stand in a continuous regime where afferents fire continuously with a constant population rate. As such, STDP is able to solve a very difficult computational problem: to localize a repeating spatio-temporal spike pattern embedded in equally dense 'distractor' spike trains. STDP thus enables some form of temporal coding, even in the absence of an explicit time reference. Given that the mechanism exposed here is simple and cheap it is hard to believe that the brain did not evolve to use it.

Introduction: The Needle in the Haystack

Neuroscientists have long observed that groups of neurons fire in precise, repeating spatio-temporal patterns. However, detecting these patterns is a significant computational challenge. In a continuous stream of neural activity, these patterns are often buried in "distractor" spikes that maintain the same average firing rate. This means the pattern is invisible to any mechanism that only looks at the frequency of spikes (rate coding).

Previous models of Spike Timing Dependent Plasticity (STDP)—the biological process where synaptic strength changes based on the millisecond-level timing of inputs—mostly assumed that spikes arrive in discrete "volleys" with a clear start time, such as a stimulus onset. This paper addresses a more complex scenario: Can a single neuron find a repeating pattern in a continuous, clock-less stream of noise where the population firing rate provides no signal of the pattern's presence?

Methods and Experimental Design

The researchers employed a computational model featuring a single Leaky Integrate-and-Fire (LIF) neuron receiving inputs from 2,000 afferents.

Input Simulation

The study simulated 450 seconds of continuous activity. For the majority of the time, afferents fired randomly according to a Poisson process. A specific 50ms spatio-temporal pattern was "copy-pasted" into the stream at random intervals. Crucially, the pattern was designed to have the same spike density as the background noise, rendering it "invisible" to rate-based detection.

Stationary population firing rate of input afferents.

The Plasticity Rule

The neuron utilized a standard STDP rule:

  • Long-Term Potentiation (LTP): Synapses are strengthened if an input spike arrives just before the neuron fires.
  • Long-Term Depression (LTD): Synapses are weakened if an input spike arrives just after the neuron fires.

Unlike previous studies, there was no external "reset" or time reference. The neuron had to autonomously identify the pattern's boundaries through the statistical repetition of spike timings.

Key Insights: Tracking the Pattern

The paper's most significant finding is the "tracking back" mechanism. Initially, the neuron might fire at a random point within the 50ms pattern. Because STDP reinforces spikes occurring immediately before the output spike, the "winning" synapses are those that fire slightly earlier. Over time, this causes the neuron's response to migrate backward through the pattern until it reaches the very beginning—the earliest point where enough coincident spikes exist to reach the threshold.

The LIF neuron acts as a coincidence detector rather than a simple integrator. It identifies spikes that arrive nearly simultaneously. STDP prunes away "noise" synapses and retains only those that consistently contribute to this early coincidence, resulting in a bimodal (binary) weight distribution where synapses in the early pattern are maximally strengthened and others are fully depressed.

Results

  • High Success & Precision: The neuron successfully learned the pattern in 96% of baseline cases.
  • Rapid Convergence: Selectivity emerged after approximately 70 pattern presentations (~13.5 seconds of simulated time), with full convergence after ~2,000 discharges.
  • Zero False Alarms: Once converged, the neuron achieved a 99.1% hit rate with no false positives, despite the dense noise.
  • Minimal Latency: The neuron eventually fired only 4ms after the start of the 50ms pattern, acting as an "earliest predictor."
  • Robustness: The system remained effective with up to 3ms of jitter and could learn even with 10% of pattern spikes missing.

Limitations and Future Directions

The current model focuses on excitatory synapses, meaning it can learn "positive" patterns (coincident spikes) but not "negative" ones (the absence of a spike). Additionally, while this study focuses on a single neuron learning one pattern, real-world biological systems involve populations of neurons. The authors suggest that inhibitory horizontal connections could allow a population to "divide and conquer," with different neurons locking onto different repeating patterns. Future work may also explore how STDP handles patterns that slowly evolve or drift over time.

Summary updated

Abstract

Experimental studies have observed Long Term synaptic Potentiation (LTP) when a presynaptic neuron fires shortly before a postsynaptic neuron, and Long Term Depression (LTD) when the presynaptic neuron fires shortly after, a phenomenon known as Spike Timing Dependent Plasticity (STDP). When a neuron is presented successively with discrete volleys of input spikes STDP has been shown to learn 'early spike patterns', that is to concentrate synaptic weights on afferents that consistently fire early, with the result that the postsynaptic spike latency decreases, until it reaches a minimal and stable value. Here, we show that these results still stand in a continuous regime where afferents fire continuously with a constant population rate. As such, STDP is able to solve a very difficult computational problem: to localize a repeating spatio-temporal spike pattern embedded in equally dense 'distractor' spike trains. STDP thus enables some form of temporal coding, even in the absence of an explicit time reference. Given that the mechanism exposed here is simple and cheap it is hard to believe that the brain did not evolve to use it.
BibTeX
@inproceedings{masquelier2008spike,
  author = {Timothée Masquelier and Rudy Guyonneau and Simon J. Thorpe},
  title = {{Spike Timing Dependent Plasticity Finds the Start of Repeating Patterns in Continuous Spike Trains}},
  booktitle = {PLoS},
  year = {2008},
  doi = {10.1371/journal.pone.0001377},
}
Figure