Solving complex learning tasks in brain-inspired computers
In a significant leap towards achieving true artificial intelligence that mirrors the efficiency of the human brain, a collaborative team of researchers from Heidelberg University and the University of Bern has made strides in training spiking neural networks. Led by Dr. Mihai Petrovici, the interdisciplinary team employed biologically-inspired artificial neural networks, specifically focusing on spiking neural networks that mimic the structure and function of the natural nervous system.
Spiking neural networks are particularly promising due to their ability to emulate the brain's powerful, fast, and energy-efficient information processing. The main distinction lies in their utilization of spike-based information processing, enabling them to efficiently handle complex tasks like image recognition and classification. Julian Göltz, a doctoral candidate in Dr. Petrovici's research group, emphasizes the efficiency of these networks, stating, "They can solve complex tasks such as image recognition and classification with extreme energy efficiency."
One of the key challenges faced by researchers is devising effective training methods for these intricate spiking neural networks. The team has successfully developed and implemented an algorithm crucial for training such networks. This algorithm ensures that neurons in the spiking neural network fire at the correct time, optimizing the network for specific tasks such as high-precision image classification.
The human brain and artificial spiking neural networks share a reliance on proper connections between individual neurons to reach their full potential. Laura Kriener, another member of Dr. Petrovici's team, underscores the importance of adjusting neuromorphic systems to process spiking input accurately. Specialized algorithms play a vital role in fine-tuning these connections, allowing the network to excel in various tasks.
The team's algorithm enables training spiking neural networks to code and transmit information exclusively in single spikes, resulting in rapid and efficient outcomes. Furthermore, the researchers successfully implemented a neural network trained with this algorithm on the BrainScaleS-2 neuromorphic hardware platform developed at Heidelberg University. According to the researchers, this system processes information up to a thousand times faster than the human brain while consuming significantly less energy than traditional computer systems.
The BrainScaleS system is part of the European Human Brain Project, contributing to the integration of neuromorphic computing into an open platform called EBRAINS. Dr. Petrovici highlights the broader implications of their work, stating, "Our work is not only interesting for neuromorphic computing and biologically inspired hardware. It also acknowledges the demand from the scientific community to transfer so-called Deep Learning approaches to neuroscience and thereby further unveil the secrets of the human brain."
This groundbreaking research, funded by the Manfred Stärk Foundation and the Human Brain Project, marks a significant step towards advancing artificial intelligence and understanding the intricacies of the human brain. The results of their work were published in the prestigious journal "Nature Machine Intelligence."
Story Source:
Materials provided by Heidelberg University. Note: Content may be edited for style and length.