This laboratory investigates how principles of biological organisation can inform the design of computing systems. We develop spiking networks, reservoir models and learning architectures whose behaviour depends on their internal dynamics and connectivity. Our work examines excitation, inhibition and memory, together with wiring rules inspired by genetic organisation and spatial structure. We test how these choices affect learning and information processing, and how computation can remain robust when device properties vary. Through neural-interface projects, the lab connects these models to adaptive control and the physical constraints of neuromorphic hardware.
Research connections
Laboratory members
Additional laboratory members (1)
Publications
2026
Entropy (Basel, Switzerland)
Recurrent neural networks are studied along two largely parallel tracks: as machine-learning models evaluated by task performance and as computational-neuroscience models of cortical circuits evaluated by dynamical realism. Reservoir computing offers a meeting point, yet the link between dynamical regime and computational performance has not been systematica…Read the full abstractCollapse abstract
Recurrent neural networks are studied along two largely parallel tracks: as machine-learning models evaluated by task performance and as computational-neuroscience models of cortical circuits evaluated by dynamical realism. Reservoir computing offers a meeting point, yet the link between dynamical regime and computational performance has not been systematically mapped in biologically constrained spiking architectures. We treat the Brunel balanced excitatory-inhibitory network as a reservoir and characterize separation capacity (kernel quality) and transient memory (corrected linear memory capacity, validated by non-parametric mutual information) across the full phase diagram. The analysis uses a four-state Markov source whose Shannon entropy rate is set in closed form by a single parameter at fixed marginal entropy. Both capabilities increase monotonically with the inhibitory ratio g, remaining jointly highest in the asynchronous irregular regime, with diminishing increments consistent with eventual saturation; the synchronous irregular regime, despite a network timescale three orders of magnitude longer, supports neither. Memory further requires sparse input coupling: dense coupling collapses the driven timescale and erases memory in every regime. Inhibitory balance thus emerges as a unified architectural control parameter, providing a quantitative design criterion for cortical-circuit modeling and reservoir computing applications.
2026
Lecture Notes in Computer Science
The high computational complexity and increasing parameter counts of deep neural networks pose significant challenges for deployment in resource-constrained environments, such as edge devices or real-time systems. To address this, we propose a parameter-efficient neural architecture where neurons are embedded in Euclidean space. During training, their positi…Read the full abstractCollapse abstract
The high computational complexity and increasing parameter counts of deep neural networks pose significant challenges for deployment in resource-constrained environments, such as edge devices or real-time systems. To address this, we propose a parameter-efficient neural architecture where neurons are embedded in Euclidean space. During training, their positions are optimized and synaptic weights are determined as the inverse of the spatial distance between connected neurons. These distance-dependent wiring rules replace traditional learnable weight matrices and significantly reduce the number of parameters while introducing a biologically inspired inductive bias: connection strength decreases with spatial distance, reflecting the brain’s embedding in three-dimensional space where connections tend to minimize wiring length. We validate this approach for both multi-layer perceptrons and spiking neural networks. Through a series of experiments, we demonstrate that these spatially embedded neural networks achieve a performance competitive with conventional architectures on the MNIST dataset. Additionally, the models maintain performance even at pruning rates exceeding 80% sparsity, outperforming traditional networks with the same number of parameters under similar conditions. Finally, the spatial embedding framework offers an intuitive visualization of the network structure.
2025
Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference
Bio-inspired networks offer rich dynamic capabilities with minimal energy demands, especially when implemented on neuromorphic hardware. In particular, the recurrence in brain circuits enables Recurrent Spiking Neural Networks (RSNNs) to generate complex spatio-temporal spike patterns, forming internal representations of time-varying signals. Despite this bi…Read the full abstractCollapse abstract
Bio-inspired networks offer rich dynamic capabilities with minimal energy demands, especially when implemented on neuromorphic hardware. In particular, the recurrence in brain circuits enables Recurrent Spiking Neural Networks (RSNNs) to generate complex spatio-temporal spike patterns, forming internal representations of time-varying signals. Despite this biological sophistication, such architectures are often applied to machine learning tasks with limited biological relevance. In this work, we present a spiking reservoir computing architecture, implemented as a fully spiking Liquid State Machine (LSM) with Leaky Integrate-and-Fire (LIF) neurons, designed to recognize and decode internal brain states. We simulate epileptic activity using a spiking model and develop a complete pipeline where a source generates Local Field Potentials (LFPs). These signals are encoded through population coding and processed by the LSM, which performs regression on the biophysical parameters controlling epileptic dynamics, thereby inferring the source state. To enhance the LSM's performance, we apply a biologically inspired synaptic plasticity mechanism to the RSNN. Our results demonstrate that a simple, unsupervised plasticity mechanism can optimize the internal parameters of the reservoir, particularly in smaller networks. This approach offers a hardware-efficient strategy for task-specific adaptation of general-purpose circuits, highlighting its suitability for edge-device implementations. Our findings emphasize the potential of spiking reservoir computing for real-time decoding of complex brain dynamics, such as epileptic activity, and underscore the advantages of biologically inspired, energy-efficient methods for neuromorphic systems in real-world applications.Clinical relevance- The ability to decode internal brain states from Local Field Potentials (LFPs) using a biologically inspired spiking reservoir computing architecture has significant implications for clinical neuroscience. By providing a real-time, energy-efficient method for tracking epileptic dynamics, this approach could aid in the development of advanced braincomputer interfaces (BCIs) and personalized neurostimulation therapies. Such a system may help clinicians monitor seizure progression, optimize treatment strategies, and improve patient outcomes in clinical epilepsy management.
2024
arXiv
Mixed-signal implementations of SNNs offer a promising solution to edge computing applications that require low-power and compact embedded processing systems. However, device mismatch in the analog circuits of these neuromorphic processors poses a significant challenge to the deployment of robust processing in these systems. Here we introduce a novel archite…Read the full abstractCollapse abstract
Mixed-signal implementations of SNNs offer a promising solution to edge computing applications that require low-power and compact embedded processing systems. However, device mismatch in the analog circuits of these neuromorphic processors poses a significant challenge to the deployment of robust processing in these systems. Here we introduce a novel architectural solution inspired by biological development to address this issue. Specifically we propose to implement architectures that incorporate network motifs found in developed brains through a differentiable re-parameterization of weight matrices based on gene expression patterns and genetic rules. Thanks to the gradient descent optimization compatibility of the method proposed, we can apply the robustness of biological neural development to neuromorphic computing. To validate this approach we benchmark it using the Yin-Yang classification dataset, and compare its performance with that of standard multilayer perceptrons trained with state-of-the-art hardware-aware training method. Our results demonstrate that the proposed method mitigates mismatch-induced noise without requiring precise device mismatch measurements, effectively outperforming alternative hardware-aware techniques proposed in the literature, and providing a more general solution for improving the robustness of SNNs in neuromorphic hardware.
