Two-compartment neuronal spiking model expressing brain-state specific apical-amplification, -isolation and -drive regimes

TL;DR

This work introduces a two-compartment Ca-AdEx neuron that captures brain-state-specific apical mechanisms—apical amplification, isolation, and drive—via a Ca^{2+}-spike in the distal compartment and an AdEx soma. A data-driven, population-based optimization (Learning-to-Learn) discovers the genome parameters that realize these regimes, with ThetaPlanes providing a compact, piecewise-linear transfer function in the somatic–distal input space. The Ca-AdEx model is implemented in a multi-compartment NEST framework and demonstrated on both simplified and extended morphologies, including NMDA-spike subunits, enabling simulations of wakefulness, NREM, and REM–like network dynamics. The approach combines biologically grounded dendritic non-linearities with HPC-driven meta-optimization to support brain-state–aware learning and offers a scalable gate for AI systems with state-dependent processing constraints.

Abstract

Mounting experimental evidence suggests that brain-state-specific neural mechanisms, supported by connectomic architectures, play a crucial role in integrating past and contextual knowledge with the current, incoming flow of evidence (e.g., from sensory systems). These mechanisms operate across multiple spatial and temporal scales, necessitating dedicated support at the levels of individual neurons and synapses. A notable feature within the neocortex is the structure of large, deep pyramidal neurons, which exhibit a distinctive separation between an apical dendritic compartment and a basal dendritic/perisomatic compartment. This separation is characterized by distinct patterns of incoming connections and brain-state-specific activation mechanisms, namely, apical amplification, isolation, and drive, which are associated with wakefulness, deeper NREM sleep stages, and REM sleep, respectively. The cognitive roles of apical mechanisms have been demonstrated in behaving animals. In contrast, classical models of learning in spiking networks are based on single-compartment neurons, lacking the ability to describe the integration of apical and basal/somatic information. This work aims to provide the computational community with a two-compartment spiking neuron model that incorporates features essential for supporting brain-state-specific learning. This model includes a piece-wise linear transfer function (ThetaPlanes) at the highest abstraction level, making it suitable for use in large-scale bio-inspired artificial intelligence systems. A machine learning evolutionary algorithm, guided by a set of fitness functions, selected the parameters that define neurons expressing the desired apical mechanisms.

Figures (9)

• Figure 1: The two-loop scheme of L2L. In the inner loop, a model is trained or simulated on a task from a family of tasks. A fitness function evaluates the performance of the model. The model parameters are optimized in the outer loop. Image provided by Yegenoglu2022exploring.
• Figure 2: Search for approximating planes. a) Representation of $\nu(I_s,I_d)$, the firing rate of the two-compartment Ca-AdEx spiking neuron in response to combinations of somatic ($I_s$) and distal ($I_d$) currents. b) Algorithmic identification of $M_{+}$ and $M_{-}$ regions from spiking simulation results.
• Figure 3: Errors of fitting planes (Hz). Panel a) $M_{+}$ region: $\nu_{+}-\nu$. b) $M_{-}$ region: $\nu_{-}-\nu$.
• Figure 4: Linearity of the separation between the high activity $M_{+}$ region and the lower activity $M_{-}$ region. (red line): $I_{d,F}^H(I_s)$ linear fit.
• Figure 5: Proxies for ACh and NA modulation. Inducing a range of apical-amplification -isolation and -drive like configurations from the same starting neuron. Parameters in table \ref{['tab:ApicalRegimes-9-configurations']}