arXiv subjects

Janina Hesse

Publications and source records attributed to Janina Hesse.

5 recordsLinked to original sources

arXiv2026

Modelling chronic stress as excitation-inhibition perturbation in recurrent working-memory networks

Chronic stress can turn an organism's stress response maladaptive, with consequences including behavioral inflexibility and impaired working memory. At the cellular level, chronic stress shifts the excitatory-inhibitory (E/I) balance of prefrontal pyramidal neurons toward inhibitory dominance, yet the mechanisms underlying these alterations, as well as their effect on network performance, are still unknown. We use recurrent E/I networks to isolate a chronic-stress-associated increase in inhibition and examine its consequences for inhibitory dominance, excitatory hypofunction, and computation. We then ask how stress adaptation preserves function and what computational costs adaptation entails. We model the effects of stress as selective stochastic strengthening of inhibitory synapses onto excitatory neurons. Applying this stress perturbation to neural networks trained on a working-memory task, task performance drops, as would be expected under chronic stress. In contrast, networks trained under strengthened inhibition preserve task performance and develop inhibitory dominance. These resilient networks stabilise recurrent dynamics under stress and an energetic proxy, and produce a sparser, less reciprocally connected circuit. These stress adaptations, however, decrease performance of resilient networks compared to networks trained without stress for tasks requiring longer working-memory. This resilience-generalisation trade-off persists across network size and stress magnitude used during training. Comparing eight candidate perturbations on synaptic strength or neuronal excitability confirms that strengthened inhibition best reproduces the signatures of stress among those tested. Our results suggest that adaptation on the network level can preserve familiar computations under stress, yet stress adaptation specialises circuits such that flexibility outside the trained regime is limited.

q-bio.NC↗

arXiv2026

What is Missing? Explaining Neurons Activated by Absent Concepts

Explainable artificial intelligence (XAI) aims to provide human-interpretable insights into the behavior of deep neural networks (DNNs), typically by estimating a simplified causal structure of the model. In existing work, this causal structure often includes relationships where the presence of a concept is associated with a strong activation of a neuron. For example, attribution methods primarily identify input pixels that contribute most to a prediction, and feature visualization methods reveal inputs that cause high activation of a target neuron - the former implicitly assuming that the relevant information resides in the input, and the latter that neurons encode the presence of concepts. However, a largely overlooked type of causal relationship is that of encoded absences, where the absence of a concept increases neural activation. In this work, we show that such missing but relevant concepts are common and that mainstream XAI methods struggle to reveal them when applied in their standard form. To address this, we propose two simple extensions to attribution and feature visualization techniques that uncover encoded absences. Across experiments, we show how mainstream XAI methods can be used to reveal and explain encoded absences, how ImageNet models exploit them, and that debiasing can be improved when considering them.

cs.CV↗

arXiv2020

Firing statistics in the bistable regime of neurons with homoclinic spike generation

Neuronal voltage dynamics of regularly firing neurons typically has one stable attractor: either a fixed point (like in the subthreshold regime) or a limit cycle that defines the tonic firing of action potentials (in the suprathreshold regime). In two of the three spike onset bifurcation sequences that are known to give rise to all-or-none type action potentials, however, the resting-state fixpoint and limit cycle spiking can coexist in an intermediate regime, resulting in bistable dynamics. Here, noise can induce switches between the attractors, i.e., between rest and spiking, and thus increase the variability of the spike train compared to neurons with only one stable attractor. Qualitative features of the resulting spike statistics depend on the spike onset bifurcations. This study focuses on the creation of the spiking limit cycle via the saddle-homoclinic orbit (HOM) bifurcation and derives interspike interval (ISI) densities for a conductance-based neuron model in the bistable regime. The ISI densities of bistable homoclinic neurons are found to be unimodal yet distinct from the inverse Gaussian distribution associated with the saddle-node-on-invariant-cycle (SNIC) bifurcation. It is demonstrated that for the HOM bifurcation the transition between rest and spiking is mainly determined along the downstroke of the action potential -- a dynamical feature that is not captured by the commonly used reset neuron models. The deduced spike statistics can help to identify HOM dynamics in experimental data.

q-bio.NC↗

arXiv2019

How to correctly quantify neuronal phase-response curves from noisy recordings

At the level of individual neurons, various coding properties can be inferred from the input-output relationship of a cell. For small inputs, this relation is captured by the phase-response curve (PRC), which measures the effect of a small perturbation on the timing of the subsequent spike. Experimentally, however, an accurate experimental estimation of PRCs is challenging. Despite elaborate measurement efforts, experimental PRC estimates often cannot be related to those from modeling studies. In particular, experimental PRCs rarely resemble the generic PRC expected close to spike initiation, which is indicative of the underlying spike-onset bifurcation. Here, we show for conductance-based model neurons that the correspondence between theoretical and measured phase-response curve is lost when the stimuli used for the estimation are too large. In this case, the derived phase-response curve is distorted beyond recognition and takes on a generic shape that reflects the measurement protocol, but not the real neuronal dynamics. We discuss how to identify appropriate stimulus strengths for perturbation and noise-stimulation methods, which permit to estimate PRCs that reliably reflect the spike-onset bifurcation -- a task that is particularly difficult if a lower bound for the stimulus amplitude is dictated by prominent intrinsic neuronal noise.

q-bio.NC↗

arXiv2016

Qualitative changes in spike-based neural coding and synchronization at the saddle-node loop bifurcation

Information processing in the brain crucially depends on encoding properties of single neurons, with particular relevance of the spike-generation mechanism. The latter hinges upon the bifurcation type at the transition point between resting state and limit cycle spiking. Prominent qualitative changes in encoding have previously been attributed to a specific switch of such a bifurcation at the Bogdanov-Takens (BT) point. This study unveils another, highly relevant and so far underestimated transition point: the saddle-node loop bifurcation. As we show, this bifurcation turns out to induce even more drastic changes in spike-based coding than the BT transition. This result arises from a direct effect of the saddle-node loop bifurcation on the limit cycle and hence spike dynamics, in contrast to the BT bifurcation, whose immediate influence is exerted upon the subthreshold dynamics and hence only indirectly relates to spiking. We specifically demonstrate that the saddle-node loop bifurcation (i) ubiquitously occurs in planar neuron models with a saddle-node on invariant cycle onset bifurcation, and (ii) results in a symmetry breaking of the system's phase-response curve. The latter entails close to optimal coding and synchronization properties in event-based information processing units, such as neurons. The saddle-node loop bifurcation leads to a peak in synchronization range and provides an attractive mechanism for the so far unresolved facilitation of high frequencies in neuronal processing. The derived bifurcation structure is of interest in any system for which a relaxation limit is admissible, such as Josephson junctions and chemical oscillators. On the experimental side, our theory applies to optical stimulation of nerve cells, and reveals that these techniques could manipulate a variety of information processing characteristics in nerve cells beyond pure activation.

q-bio.NC↗