According to the technology media Tom's Hardware, a research team from Tohoku University and Future University in Japan recently successfully trained rat cortical neurons, using a real-time machine learning framework that can autonomously generate complex temporal signals. This breakthrough research opens up new possibilities for the application of neurons in AI computing.
The research team combined living neurons with high-density microelectrode arrays and microfluidic devices to build a system called "closed-loop reservoir computing." The unique feature of this system is that it can learn and generate periodic and chaotic waveforms without external input, thereby performing various AI computing tasks.
The team's core technology involves using polydimethylsiloxane (PDMS) microfluidic films to constrain connections between neurons. Studies show that without physical constraints, neurons form highly synchronized networks, which are not effective for learning target signals. To solve this problem, researchers confined neuronal cell bodies in 128 micro-pores, connected by microchannels, forming two types of network structures: grid and hierarchical. This design significantly enhanced the dynamical dimension of the network, reduced the correlation between neurons, and improved system performance.
In testing, the grid network demonstrated excellent performance, capable of generating sine, triangle, and square waves with periods of 4 seconds, 10 seconds, and 30 seconds, and even approximating three-dimensional chaotic trajectories - the Lorenz attractor. During the learning phase, the system's predicted signal had a correlation of over 0.8 with the target signal, showing good learning ability. Professor Hidemasa Yamamoto from Tohoku University stated that living neuronal networks have biological significance and can also serve as a new type of computational resource.
Although some results have been achieved, the team still faces some technical challenges, especially in performance. The study shows that after training stops, the system's error increases during autonomous operation, and a 330-millisecond feedback loop delay limits the system's ability to track rapidly changing waveforms. In the future, the research team hopes to develop specialized hardware to reduce the delay, thus expanding the application prospects of this technology in brain-computer interfaces and neural prosthetic devices.
Key points:
🌟 Japanese scientists successfully trained rat neurons to autonomously generate complex temporal signals.
🧠 Using microfluidic technology, a closed-loop reservoir computing system was built, which requires no external input.
⚙️ In the future, the team hopes to improve hardware to expand the application of this technology in brain-computer interfaces and neural prosthetics.
