Chinese Scientists Develop New AI That Can Imitulate Brain Neurons

Chinese scientists are developing AI models with new infrastructure (photo: dock. unsplash)

JAKARTA A team of researchers from China developed a new Artificial Intelligence (AI) system with a unique concept. This AI system mimics human brain neurons that can reduce energy needs.

The AI system, named SpikingBrain-1.0, was developed by scientists from the Institute of Automation under the Chinese Academy of Sciences. This large-scale model is trained using domestically made GPU computing.

In contrast to a mainstream generative AI model like ChatGPT that uses a Transformer architecture, SpikingBrain-1.0 takes a different path. SpikingBrain-1.0 allows intelligence to emerge from soaring neurons.

Therefore, this AI model does not require a lot of resources. SpikingBrain-1.0 also shows extraordinary training efficiency. The model uses only 2 percent of the pre-training data that large models need in general.

Although the training data is less, this model offers a performance that is comparable to the open source model, especially in terms of language understanding and reasoning. In addition, SpikingBrain-1.0 also offers a significant increase in speed.

By exploiting a spike in neurons, one variant of this AI model is shown to be 26.5 times faster than the AI model that uses Transformers. This can be seen when both models are tested to produce the first token from the context of one million tokens.

Therefore, SpikingBrain-1.0 can handle a variety of complex tasks, including analyzing legal and medical documents. Moreover, these AI models can also be used in particle physics experiments and DNA-series modeling.

The research team has made the SpikingBrain model an open source and launched it on a public test page. This move will allow other researchers to collaborate and develop the technology.