Spike-timing-dependent plasticity and temporal input statistics

Henning Sprekeler, Laurenz WiskottBMC 2007 · DOI

Abstract

Recent research has focused on the functional interpretations of spike-timing-dependent plasticity (STDP). This study utilizes a linear Poisson neuron model to analytically examine how temporal input statistics influence synaptic weight distributions. We demonstrate that the outcome of learning is determined by an "effective learning window"—the convolution of the STDP learning window with the excitatory post-synaptic potential (EPSP) shape. This perspective suggests that STDP acts as a frequency filter, potentially inverting the low-pass filtering of the EPSP or acting as a band-pass filter linked to cortical rhythms. Furthermore, we find that causal (irreversible) input statistics can destabilize weight distributions, challenging the traditional view of STDP as a simple causality detector.

Introduction

Spike-Timing-Dependent Plasticity (STDP) is a fundamental biological process where synaptic strength is modified based on the precise millisecond-level timing of pre- and post-synaptic spikes. Traditionally, the "learning window"—the curve defining potentiation or depression based on spike timing—has been studied in isolation. However, neurons possess physical properties, notably the Excitatory Post-Synaptic Potential (EPSP), which describes the temporary change in membrane potential following a signal. This paper addresses the gap in understanding how the temporal statistics of input signals interact with both the STDP rule and the physical shape of the EPSP to determine synaptic weight distributions.

Methods: The Effective Learning Window

The study employs an analytical approach using a linear Poisson neuron model. The central innovation is the introduction of the "effective learning window." This is mathematically defined as the convolution of the standard STDP learning window with the shape of the EPSP. By analyzing this effective window, we can predict neuronal responses to various input statistics, such as signals varying slowly over time or those exhibiting sharp causal dependencies. This framework allows for a more comprehensive understanding of how a neuron's physical response characteristics modulate the learning process.

Key Insights and Results

STDP as a Frequency Filter

The interaction between the STDP rule and the EPSP suggests that the EPSP acts as a filter. A significant insight is that the asymmetric STDP window (where "pre-before-post" causes potentiation and "post-before-pre" causes depression) may serve to counteract the low-pass filtering effect inherent in the EPSP. By "sharpening" the signal that the neuron's physics naturally "blurs," STDP effectively inverts the low-pass filter.

Furthermore, STDP can act as a frequency selector:

  • Low-pass filtering: If the effective learning window is symmetric and low-pass, it focuses on slowly varying signals. This is crucial for learning invariant sensory representations, such as recognizing an object despite movement.
  • Band-pass filtering: In cases where the EPSP is short, the effective window can act as a band-pass filter. This suggests a potential link between STDP and specific cortical rhythms, such as theta or gamma oscillations.

Challenging the Causality Detector Hypothesis

A prevalent theory in neuroscience posits that STDP is a mechanism for detecting causal relationships. However, our analytical results show that "irreversible" (causal) input statistics tend to destabilize synaptic weights, leading to oscillations. This finding suggests that the asymmetric learning window may serve a different functional purpose than simple causality detection.

Functional Equivalence

The analysis proves that different STDP learning windows can yield identical functional outcomes if the EPSP shapes are appropriately adjusted. This implies that learning rules observed across different brain regions might be functionally more similar than their diverse physical shapes initially suggest.

Conclusion and Future Directions

This research shifts the focus from the isolated STDP learning window to the integrated "effective learning window." While the linear Poisson model provides elegant analytical solutions, it is limited by its linearity. Real neurons exhibit complex non-linear behaviors, such as refractory periods and bursting. Future research should explore these findings in more complex, non-linear models and further investigate the connection between short EPSPs, STDP, and the tuning of neurons to specific cortical rhythms.

Summary updated

Abstract

A lot of research has recently focused on possible functional interpretations of the peculiar dependence of synaptic plasticity on the relative timing of pre- and post-synaptic spikes. Here we use a linear Poisson neuron to analytically examine how the temporal statistics of the input signals influence the distribution of the synaptic weights. The analysis shows that the outcome of learning is not determined by the shape of the learning window alone, but rather by the convolution of the learning window with the shape of the excitatory post-synaptic potential (EPSP), subsequently referred to as the effective learning window. This indicates that very different learning windows may have the same functional role depending on the shape of the EPSP. Moreover, it offers a new interpretation of the commonly observed asymmetry of the learning window of spike-timing-dependent plasticity (STDP) as a mechanism for inverting neuronal low-pass filtering as invoked by the EPSP. For reversible input statistics, the learning rule shows a preference for certain frequency ranges in the input signals. If the symmetric component of the effective learning window has the form of a low-pass filter, STDP focuses on low-frequency components of the input signals, i.e., components that vary slowly relative to intrinsic time scales given by the learning window and the EPSP. This is in line with several learning paradigms that have been proposed as mechanisms for learning invariant sensory representations and for the self-organized formation of visual receptive fields. Moreover, in case the EPSP is short, the effective learning window acts as a band pass filter, leading to the speculation that there could be a connection between cortical rhythms and STDP learning. Interestingly, it turns out that irreversible input statistics, e.g., causal dependencies between the input signals, tend to destabilize the weight distribution and favor oscillating weights. This observation challenges the interpretation of the asymmetric learning window as a causality detector.
BibTeX
@inproceedings{sprekeler2007spike,
  author = {Henning Sprekeler and Laurenz Wiskott},
  title = {{Spike-timing-dependent plasticity and temporal input statistics}},
  booktitle = {BMC},
  year = {2007},
  doi = {10.1186/1471-2202-8-s2-p86},
}
Figure