.png?2.1.1 "Vokash")
9 hours ago
A investigation squad astatine Tohoku University and Future University Hakodate has demonstrated that surviving biologic neurons tin beryllium trained to execute a supervised temporal shape learning task antecedently carried retired by artificial systems. By integrating cultured neuronal networks into a instrumentality learning framework, nan squad showed that these biologic systems tin make analyzable time-series signals, marking a important measurement guardant successful some neuroscience and bio-inspired computing.
The study was published online successful Proceedings of nan National Academy of Sciences (PNAS) connected March 12, 2026, highlighting a caller intersection betwixt surviving neural systems and computational technology. The findings propose that biologic neural networks (BNNs) whitethorn service arsenic viable alternatives aliases complements to existing instrumentality learning models.
Artificial neural networks (ANNs) and spiking neural networks (SNNs) person agelong been utilized successful instrumentality learning and neuromorphic hardware. A model known arsenic reservoir computing has emerged arsenic an businesslike attack for processing time-dependent information by leveraging nan move properties of recurrently connected ANNs and SNNs.
In accepted ANN-based reservoir computing, methods specified arsenic First-Order Reduced and Controlled Error (FORCE) learning alteration real-time adjustment by continuously adjusting output signals successful consequence to errors. These techniques let artificial systems to make a wide scope of temporal patterns, including periodic and chaotic signals. However, whether akin approaches could beryllium applied to biologic neural networks has remained an unfastened question.
To reside this gap, nan researchers constructed biologic neural networks utilizing cultured rat cortical neurons and incorporated them into a reservoir computing framework. By applying FORCE learning to optimize nan system's readout layer, nan squad successfully trained nan biologic networks to nutrient analyzable temporal signals comparable to those progressive successful centrifugal control.
A cardinal invention successful nan study was nan usage of microfluidic devices to precisely guideline neuronal maturation and power web connectivity. This attack enabled nan researchers to create modular web architectures that minimized excessive synchronization, thereby promoting nan rich, high-dimensional dynamics required for effective reservoir computing.
Using this system, nan BNN-based model was capable to make a assortment of time-series patterns, including sine waves, triangular waves, quadrate waves, and moreover chaotic trajectories specified arsenic nan Lorenz attractor. Notably, nan web demonstrated elasticity by learning and stably reproducing sine waves pinch periods ranging from 4 to 30 seconds wrong nan aforesaid system.
This activity shows that surviving neuronal networks are not only biologically meaningful systems but whitethorn besides service arsenic caller computational resources. By bridging neuroscience and instrumentality learning, we are opening a pathway toward caller forms of computing that leverage nan intrinsic dynamics of biologic systems."
Hideaki Yamamoto, professor astatine Tohoku University
Looking ahead, nan investigation squad intends to amended nan stableness of awesome procreation aft training has concluded. Future efforts will attraction connected reducing feedback delays and refining nan FORCE learning algorithm. In parallel, nan level whitethorn beryllium expanded into a microphysiological strategy for studying supplier responses and modeling neurological disorders, further extending its effect crossed some technological and aesculapian fields.
Source:
Journal reference:
Sono, Y., et al. (2026). Online supervised learning of temporal patterns successful biologic neural networks nether feedback control. Proceedings of nan National Academy of Sciences. DOI: 10.1073/pnas.2521560123. https://www.pnas.org/doi/10.1073/pnas.2521560123
English (US) · 
Indonesian (ID) ·