Adaptive learning in cortical networks

: a computational model of Cholinergic neuromodulation

Abstract

The cholinergic system is essential for learning, attention, memory, and other cognitive functions. Acetylcholine (ACh) has been linked to multiple forms of synaptic plasticity, including experience-dependent plasticity and the expansion of receptive fields for behaviorally relevant stimuli. It is also believed to play a key role in adaptive plasticity. The importance of ACh in maintaining these functions is reflected in its association with cognitive decline. Evidence suggests that ACh enhances resilience to neural damage, providing neuroprotective effects in the presence of lesions, aging, and neurodegenerative disorders. Cholinergic dysfunction is a hallmark of cognitive impairment, with basal forebrain degeneration serving as a key pathological marker in Alzheimer’s Disease (AD).
Despite extensive research, computational frameworks that explain how cholinergic modulation can both enhance learning and contribute to cognitive dysfunction remain scarce. This is largely due to the complexity of the effects of ACh in the cortex, which depend on receptor type, local circuit dynamics, and the unclear spatial specificity of cholinergic modulation. As a result, defining a singular role for ACh in the cortex remains a challenge.
Artificial neural networks (ANNs) offer a powerful tool for studying these dynamics, allowing for controlled manipulation of low-level features to examine high-level behavioral effects. The work presented here employs biologically plausible neural network models and adaptive learning algorithms to investigate how adaptive learning rate modulation influences learning, neural representations, and network resilience under neurodegeneration. Although the involvement of ACh in AD is complex, this work explores whether some of the cognitive deficits in AD can be explained by impaired cholinergic modulation and the resulting disruptions in plasticity and neural representations. In the proposed model, learning is modulated using the history of error signals in cortical ensembles, linking it to cholinergic responses to novel or unexpected stimuli. Additionally, this work examines the spatial specificity of cholinergic modulation, specifically whether local circuit modulation is necessary to preserve both its neuroprotective and learning enhancing effects.
These findings suggest that cholinergic control of local cortical learning may underlie rapid learning in the cortex while greatly increasing robustness to perturbations. By linking cholinergic modulation to both learning and cognitive decline, this work provides a novel theoretical framework to understand its role in health and disease.

Date of Award	17 Jun 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	E J Coulthard (Supervisor) & Rui Ponte Costa (Supervisor)