Biologically plausible learning in spiking neural networks

Spiking Neural Networks (SNNs) represent the "third generation" of neural models, capturing the discrete, asynchronous, and energy-efficient nature of biological computation. Unlike conventional artificial neural networks, which rely heavily on backpropagation, a process considered biologically implausible due to its requirement for global weight information and precise derivative calculation, SNNs offer a platform to explore how the brain actually learns. By mimicking the spike-based communication of the mammalian cortex, SNNs provide unique insights into temporal coding and event-driven processing. However, training these networks remains a significant hurdle. Traditional gradient-based methods are difficult to apply to non-differentiable spikes, leading researchers to investigate alternative learning rules like Spike-Timing-Dependent Plasticity (STDP), homeostatic scaling, and three-factor rules. Understanding these biologically grounded mechanisms is essential for both advancing neuroscience and developing next-generation neuromorphic hardware.The central challenge in this field is the "credit assignment problem": how can a neural network determine which specific synapses are responsible for a global outcome using only locally available information? While backpropagation excels in artificial environments, biological systems solve this through local plasticity, predictive coding, and neuromodulatory feedback loops. The goal of this Research Topic is to address the disconnect between high-performance AI and biological plausibility by identifying learning algorithms that operate under the constraints of the nervous system. We seek to showcase research that implements local learning rules within multi-layered spiking architectures. A key objective is to determine how SNNs can achieve competitive performance on complex tasks, such as pattern recognition, motor control, or temporal sequence learning, without violating biological principles like weight symmetry and non-local gradients. By fostering a dialogue between computational modeling and experimental neurobiology, this collection aims to accelerate the development of SNNs that are not only theoretically sound but also practically viable for ultra-low-power neuromorphic engineering.This Research Topic focuses on the theoretical development and practical application of learning mechanisms in Spiking Neural Networks that adhere to biological principles. We welcome contributions that explore the mathematical foundations of local plasticity, the role of temporal dynamics in learning, and the scaling of these models to complex functional tasks.Specific themes we encourage contributors to address include:- Hebbian learning and variants of Spike-Timing-Dependent Plasticity (STDP) in deep spiking architectures.- Biologically plausible approximations of backpropagation (e.g., feedback alignment, target propagation, and surrogate gradients).- The role of neuromodulators (e.g., dopamine, noradrenaline) as third factors in local learning rules.- Homeostatic mechanisms and intrinsic plasticity for stabilizing network dynamics during learning.- Predictive coding and error-driven learning frameworks in spiking hierarchies.- Implementation of biologically plausible rules on neuromorphic hardware (e.g., Loihi, TrueNorth, or SpiNNaker).- Learning in SNNs for event-based sensory processing and real-time motor control. Topic editors: Arash Ahmadi, Chris G. Antonopoulos Source: https://www.frontiersin.org/research-topics/81084/biologically-plausible-learning-in-spiking-neural-networks