summary: The lines between biology and computer science are becoming more blurred. Researchers have successfully trained neurons in living rats to perform complex machine learning tasks. In this study, we integrated cultured neuron networks into a “reservoir computing” framework.
Using a technique called FORCE learning, the researchers taught these biological circuits to generate complex mathematical patterns containing chaotic Lorentzian attractors, proving that living “wetware” can act as functional real-time computational resources.
important facts
- Reservoir computing: This framework exploits the “natural” messiness and complexity of networks (reservoirs) to process data. Rather than training every neuron, scientists only train a “readout” layer that interprets the network’s activity.
- Forced learning: A method used to adjust the output signal in real time based on error. This is the first time that it has been successfully applied. Biological Neural Network (BNN) Generate time series data.
- “Chaos” test: Living neurons have not only learned simple sine waves. they succeeded in reproducing lorenz attractora complex set of equations used to model chaotic systems like weather patterns.
- Microfluidic precision: The researchers used tiny “pipes” (microfluidics) to guide how neurons grow. By creating modular cell “neighborhoods,” they prevented all neurons from firing simultaneously (synchronization), which is important for high-level computing.
- Versatility: The same biological system is flexible enough to learn waves with periods ranging from 4 to 30 seconds, demonstrating that living networks are surprisingly adaptable.
sauce: Tohoku University
A research team from Tohoku University and Future University Hakodate has demonstrated that living biological neurons can be trained to perform supervised temporal pattern learning tasks that were previously performed by artificial systems.
By integrating cultured neuronal networks into a machine learning framework, the research team showed that these biological systems can generate complex time-series signals, a major advance in both neuroscience and bio-inspired computing.
This study was published online Proceedings of the National Academy of Sciences (PNAS) On March 12, 2026, we focus on new intersections between living neural systems and computational technologies. This finding suggests that biological neural networks (BNNs) may serve as a viable alternative or complement to existing machine learning models.
Artificial neural networks (ANN) and spiking neural networks (SNN) have long been used in machine learning and neuromorphic hardware. A framework known as reservoir computing has emerged as an efficient approach for processing time-dependent data by exploiting the dynamic properties of recursively connected ANNs and SNNs.
In traditional ANN-based reservoir computing, methods such as first-order reduced control error (FORCE) learning enable real-time adaptation by continuously adjusting the output signal in response to errors.
These techniques allow artificial systems to generate a wide range of temporal patterns, including periodic and chaotic signals. However, whether a similar approach can be applied to biological neural networks remains an open question.
To address this gap, the researchers used cultured rat cortical neurons to build a biological neural network and incorporated it into a reservoir computing framework.
By applying FORCE learning to optimize the system’s readout layer, the researchers were able to train a biological network to generate complex temporal signals comparable to those involved in motor control.
A key innovation in this study was the use of microfluidic devices to precisely guide neuron growth and control network connectivity. This approach allowed the researchers to create a modular network architecture that minimizes excessive synchronization, thereby facilitating the rich, high-dimensional dynamics required for effective reservoir computing.
Using this system, the BNN-based framework was able to generate a variety of time series patterns, including chaotic trajectories such as sinusoids, triangle waves, square waves, and even Lorentz attractors. In particular, this network demonstrated its flexibility by learning and stably reproducing sine waves with periods ranging from 4 to 30 seconds within the same system.
“This study shows that living neuron networks are not only biologically meaningful systems, but also have the potential to serve as new computational resources,” said Professor Hideaki Yamamoto of Tohoku University.
“By bridging neuroscience and machine learning, we are paving the way for new forms of computing that exploit the unique dynamics of biological systems.”
Looking ahead, the research team aims to improve the stability of signal generation after training. Future work will focus on reducing the feedback delay and improving the FORCE learning algorithm. In parallel, this platform could be extended to microphysiological systems for studying drug responses and modeling neurological disorders, further expanding its impact in both scientific and medical fields.
Answers to key questions:
answer: We are moving in that direction! this is called “Wetware Computing” Unlike traditional silicon chips, these biological reservoirs take advantage of the inherent “noisy” physics of living cells to solve problems. They are incredibly energy efficient and can adapt to new information in ways that rigid AI models often struggle with.
answer: It’s like a conductor leading an orchestra. The neuron “reservoir” is already playing millions of different sounds. Researchers are using forced learning Listen to those sounds and “reward” sounds that match the desired pattern (such as a sine wave). Over time, the output layer learns exactly which neurons to “listen” to in order to get the correct result.
answer: biology is the ultimate master parallel processing. A single biological network can process large amounts of time-dependent data with very little power. Additionally, these systems could be used to test how drugs affect “thinking” circuits or to create models of neurological diseases in a dish, without the need for animal experiments.
Editorial note:
- This article was edited by the editors of Neuroscience News.
- Journal articles were reviewed in full text.
- Additional context added by staff.
About this AI and neuroscience research news
author: Public relations office
sauce: Tohoku University
contact: Public Relations Office – Tohoku University
image: Image credited to Neuroscience News
Original research: Open access.
“Online supervised learning of temporal patterns in biological neural networks under feedback control” Yuki Sono, Hideaki Yamamoto, Yuki Nishi, Takuma Washimi, Yuya Sato, Ayumi Hirano Iwata, Yuichi Katori, Shigeo Sato. PNAS
DOI:10.1073/pnas.2521560123
abstract
Online supervised learning of temporal patterns in biological neural networks under feedback control
In vitro biological neural networks (BNNs) provide a well-defined model system to constructively investigate how living cells interact with their environment to form high-dimensional dynamics. This model system can be used to generate coherent time outputs required for motor control.
Here, by integrating cultured cortical neurons with microfluidic devices and high-density microelectrode arrays, we develop a real-time closed-loop BNN system that can generate periodic and chaotic temporal signals.
We show that by training a simple linear decoder with fixed feedback weights, the system can learn and generate different temporal patterns autonomously. When the feedback is turned on, the irregular activity of the BNN is transformed into low-dimensional structured dynamics, producing consistent trajectories characterized by stable transitions between different neural states.
The BNN trained with different target frequencies ranging from 4 to 30 seconds can be trained to maintain oscillations at different frequencies, demonstrating its adaptability. Importantly, top-down control of self-organizing network formation by microfluidic devices is key to suppress excessive synchronization, increase the dynamic complexity of BNNs, and facilitate the training process and robust output generation.
This work provides a biologically inspired platform for understanding the physical basis of cortical computation and advancing energy-efficient neuromorphic computing paradigms.
