Chinese scientists unveil artificial intelligence inspired by the human brain that outperforms traditional systems

Faced with the energy and hardware limitations of traditional models, Chinese researchers have developed a new artificial intelligence (AI) system called SpikingBrain. Inspired by the functioning of the human brain, this innovative AI is distinguished by its remarkable efficiency, achieving up to 100 times the performance of conventional systems. By leveraging spiking neural networks, it requires significantly less data for training, representing a major breakthrough in the field of artificial intelligence. Chinese researchers have developed an innovative artificial intelligence model called SpikingBrain, designed to mimic the functioning of the human brain. This model outperforms traditional large language models in terms of energy efficiency and performance. By reducing the reliance on large amounts of data and expensive graphics chips, SpikingBrain could revolutionize the use of artificial intelligence, particularly in resource-limited environments.Inspiration from the human brain for artificial intelligence The SpikingBrain model is inspired by biological neural processes. The researchers focused on how biological neurons communicate through electrical impulses. SpikingBrain’s spiking neural networks (SNNs) use a similar mechanism. This unique approach allows for the creation of much more energy-efficient systems, as the artificial « neurons » are selectively activated, firing only when needed.Reduced Data and Computing Power Requirements SpikingBrain was designed to perform better while requiring fewer resources. Traditional models, such as those used by chatbotsChatGPT and Gemini consume a lot of computing power and data. In China, researchers have had to face restrictions on access toNvidia graphics chipsused for these calculations. SpikingBrain represents a significant advancement because it was trained on less than 2% of the data typically required by a large language model, while still delivering similar performance.

Impressive performance and speed Two versions of SpikingBrain were produced, comprising 7 billion and 76 billion parameters. In tests, the version with 7 billion parameters demonstrated more than 100 times the speed of conventional models in certain configurations. On a test with a context of one million tokens, it was 26.5 times faster in generating the first token. These results show that SpikingBrain is capable of delivering exceptional performance with fewer resources.Openness to new possibilities

Beyond its performance, SpikingBrain offers interesting perspectives in terms of

shortage of content . While current resources available on the Internet are no longer sufficient for training conventional AI, SpikingBrain paves the way for a more optimized and sustainable use of data. In addition, these models are optimized to run on alternative processors, such as those from MetaX, and could become even more efficient with the advent of computing. neuromorphic.A promising future for artificial intelligence

SpikingBrain is available under a free license on GitHub, allowing researchers and developers around the world to contribute to its evolution. With similar advances, such as those undertaken by German researchers, the use of

impulses to optimize the AI is gaining popularity. This trend could herald a paradigm shift in how artificial intelligence is designed and used, reducing its energy footprint while maximizing its efficiency.