Chinese scientists have unveiled supercomputers, particularly the monkey supercomputers, built on brain-like architecture.
The system, called Darwin Monkey or “Wukong,” has over 2 billion artificial neurons and over 100 billion synapses, and is roughly equivalent to the neural structure of a macaque.
Researchers hope to serve as a simulation tool for neuroscientists while also providing a foothold towards artificial general information (AGI).
You might like it
BR (AI) N power
Unlike traditional artificial neural networks that follow classical computing principles and process data via continuous changing binary values, nervous system systems like the Darwin Monkey are driven by spike neural networks (SNNS).
SNNS mimics how signals are transmitted between neurons in the mammalian brain, and processes and transmits data via on and off bursts (or spikes) of activity in response to electrical signals.
Related: China is developing a new light-based chiplet that can move artificial general information.
Biological neurons fire electrical pulses when signals received from other neurons reach a level sufficient to trigger a response. Artificial neurons in SNNS mimic this mechanism and only fire when sufficient electrical input is constructed.
When software-based neural networks are a collection of machine learning algorithms arranged to emulate the human brain, SNNs physically replicate the way information between biological neurons moves. This configuration makes SNNs more powerful than traditional supercomputer architectures, as they allow data to be processed in parallel.
It can also be more energy efficient. Artificial neurons enter a short period of rest after each spike, during which time they are unable to respond to new inputs. This limits the frequency they fire and helps reduce overall power consumption.
Researchers say that Darwin Monkeys are powered by a 960 Darwin III neuronal chip, but each has a 960 Darwin III neuronal chip that supports up to 2.35 million picking neurons, yet consume just 2,000 watts of power.
Other Neurotype Computers
Previous record holders of neuronal computing were Intel’s Hala Point system, consisting of 115 billion artificial neurons and 128 billion artificial synapses, exceeding 140,544 processing cores.
Intel claims that its system can perform 20 times more operations per second, or 20 petaops. However, because there are so few neural forms of computers currently in existence, it is difficult to compare them in a similar way, as data is processed differently from supercomputers.
In a statement translated into English using Google Translate, the team behind Darwin Monkey said that the platform has already demonstrated the capabilities of cognitive tasks such as logical reasoning, content generation and mathematical problem solving using AI models developed by Chinese AI startup Deepseek.
The system has also been used to simulate the brains of animals with varying levels of neural complexity, including zebrafish and mice, as part of a broader effort to support brain science research.
The Darwin Monkey was designed by researchers from Zhijiang University and Zhijiang Lab. This is a research institute jointly founded by the Chinese technological conglomerate Zhijiang State Government and the Alibaba Group.
This will follow the release of Darwinmouth (“Mickey”) in September 2020. This includes 120 million artificial neurons. This corresponds to the brain of a mouse.
Source link
