Dopaminergic Wanting Modulation for Plasticity Control in a Bioinspired Taste Spiking Neural Network
Vladimir A. Lazovskiy1, Sergey V. Stasenko1,2, Ivan M. Kipelkin1,2, and Victor B. Kazantsev1,2
1 Moscow Center for Advanced Studies, 20 Kulakova Str., 123592 Moscow, Russia
2 Neurotechnology Department, Institute of Biology and Biomedicine, Lobachevsky State University of Nizhny Novgorod, Gagarin Avenue 23, 603022 Nizhny Novgorod, Russia
Introduction
Neuromorphic models of the gustatory system provide a biologically plausible framework for studying sensory information processing and adaptive learning. Although dopamine is known to regulate reinforcement learning in biological neural systems, its role as a modulatory mechanism in taste-oriented spiking neural networks (SNNs) remains insufficiently explored. This study introduces a dopaminergic "wanting" mechanism into a bioinspired Taste-SNN and evaluates its influence on learning dynamics, neural representations, and decoding performance.
Methods
The proposed architecture consists of Gaussian place cells, two layers of leaky integrate-and-fire neurons, and a Ridge regression decoder for reconstructing four-dimensional taste stimuli (sweet, sour, salty, and bitter). Learning combines unsupervised STDP, a three-factor excitatory plasticity rule, memristive inhibitory STDP, and astrocyte-inspired modulation. The novel contribution is a trial-based dopaminergic module that estimates stimulus value using a temporal-difference critic and multiplicatively gates synaptic plasticity without directly inducing weight changes. Both the baseline and dopamine-modulated networks were trained under identical five-stage curriculum learning with matched initialization and evaluation protocols.
Results
Both models converged successfully, reaching similar validation losses (~0.55) while maintaining physiologically plausible firing rates. Dopaminergic modulation increased the final mean excitatory synaptic weight (0.789 vs. 0.778) and produced a balanced distribution of potentiation and depression events. The modulation increased the average neuronal selectivity index by approximately 17%, indicating broader distribution of stimulus selectivity across the active neuronal population. Off-grid decoding performance remained essentially unchanged (R² ≈ 0.38), demonstrating that the proposed modulation preserves reconstruction accuracy while modifying internal learning dynamics. Shuffle-control experiments confirmed that decoding performance reflected genuine stimulus-specific representations rather than decoder overfitting.
Conclusions
The proposed dopaminergic wanting mechanism provides a biologically motivated extension of Taste-SNN by modulating plasticity through reinforcement-like signals while preserving network stability and decoding performance. Although no improvement in reconstruction accuracy was observed at the reduced network scale, the method enhanced internal neural organization and selectivity, suggesting promising directions for larger neuromorphic gustatory systems and reinforcement-driven sensory processing.
Acknowledgment
The work on developing the model was carried out with the support of the federal assignment of the Ministry of Science and Higher Education of the Russian Federation (project No. FSMG-2025-0070).