Abstract Cortical microcircuits are very complex networks, but they are composed of a relatively small number of stereotypical motifs. Hence one strategy for throwing light on the computational function of cortical microcircuits is to analyze emergent computational properties of these stereotypical microcircuit motifs. We are addressing here the question how spike-timing dependent plasticity (STDP) shapes the computational properties of one motif that has frequently been studied experimentally: interconnected populations of pyramidal cells and parvalbumin-positive inhibitory cells in layer 2/3. Experimental studies suggest that these inhibitory neurons exert some form of divisive inhibition on the pyramidal cells. We show that this data-based form of feedback inhibition, which is softer than that of winner-take-all models that are commonly considered in theoretical analyses, contributes to the emergence of an important computational function through STDP: The capability to disentangle superimposed firing patterns in upstream networks, and to represent their information content through a sparse assembly code.
Introduction
A fundamental challenge in neuroscience is understanding how the brain's micro-architecture supports complex computation. Traditional theoretical models often rely on "Winner-Take-All" (WTA) architectures, where a single neuron suppresses its neighbors to produce a clear output. However, biological evidence from cortical Layer 2/3 suggests a more nuanced "soft" inhibition mechanism. This study investigates how biologically realistic divisive inhibition, combined with Spike-Timing Dependent Plasticity (STDP), enables cortical motifs to perform sophisticated tasks like blind source separation and sparse coding.
Methods
The researchers developed Model $\mathcal{M}$, a data-driven microcircuit model comprising 400 excitatory (E) pyramidal neurons and 100 inhibitory (I) PV+ neurons. The model utilizes a stochastic spike response framework where firing rates depend exponentially on membrane potential. Crucially, inhibition is modeled as a divisive normalization factor rather than a simple subtractive one, aligning with in-vivo observations.
The connectivity of the model is grounded in experimental data, specifically from the Petersen Lab. The parameters for these connections, including probabilities and synaptic weights, are summarized in the table below.
Learning is driven by STDP on the synapses from input neurons to excitatory neurons. The model was tested on three benchmarks: orientation selectivity (2D pixel arrays), spatio-temporal patterns (Ornstein-Uhlenbeck processes), and Földiák’s superposition-of-bars task, which requires the network to disentangle overlapping horizontal and vertical bars.
Results and Discussion
The transition from "hard" WTA to "soft" divisive inhibition effectively transforms the circuit into a k-Winners-Take-All (k-WTA) system. In this regime, stimuli are represented by "assemblies" of approximately 17 neurons rather than single units. This distributed representation allows the network to autonomously "unmix" signals. In the Földiák task, the model achieved an F1-measure of 0.87, successfully identifying individual components within mixed inputs—a task where standard WTA models typically fail due to their inability to form modular codes.
Furthermore, the model replicated the characteristic ~3.5ms lag between excitation and inhibition observed in biological circuits. The inclusion of I-I connections was found to be critical for sharpening this temporal window, which in turn stabilizes the STDP-driven learning process.
By moving away from idealized WTA models toward biologically grounded motifs, this research demonstrates that "soft" inhibition is not a limitation but a computational feature. It prevents combinatorial explosions by allowing modular representations of complex scenes. Future work will explore the impact of inhibitory plasticity (iSTDP) and the behavior of these motifs when integrated into deeper, hierarchical architectures.
Summary updated