Monkey Neurons Inspire Pocket-Sized, Energy-Efficient AI Model
Researchers at Cold Spring Harbor Laboratory (CSHL) have developed a new artificial intelligence (AI) model inspired by the human brain, achieving high performance with a significantly smaller size and lower energy consumption than current AI systems. The model was created using data derived from neurons in macaque monkeys, according to findings published this week.
The recent surge in AI capabilities has been driven by the increasing computational power of silicon chips. But, this power comes at a substantial cost in terms of energy and water usage. Recognizing this inefficiency, researchers turned to the brain – the most efficient computational platform known – for inspiration. Despite its immense capabilities, current AI struggles to replicate the brain’s energy efficiency; the human brain operates on less power than a standard lightbulb.
The CSHL team, collaborating with Carnegie Mellon and Princeton Universities, initially aimed to understand the workings of the human brain, specifically its visual system. Their research focused on a group of neurons called V4, a component of the visual system responsible for encoding information such as color, texture, and shapes. By analyzing the activity of these neurons in macaques, the researchers were able to develop their new AI model.
A key innovation was a technique to compress the data and eliminate redundancies within the model. This resulted in a file size modest enough to be sent as an email attachment, a stark contrast to the massive data requirements of contemporary AI systems. This reduced size allows researchers to observe the model’s internal workings, a feat impossible with larger, more complex AI models.
“We found that some V4 neurons respond to shapes with sharp edges, although others respond to small dots within an image,” explained Ben Cowley, an assistant professor at CSHL involved in the research. “If our brains can accomplish more with less complex models than these AI systems, it tells us something important about the limitations of current AI approaches.”
According to CSHL, the laboratory is a private, non-profit institution with research programs focusing on cancer, neuroscience, botany, genomics, and quantitative biology. It is located in Laurel Hollow, New York, and has been an NCI-designated Cancer Center since 1987. Founded in 1890, CSHL has played a central role in the development of molecular genetics and molecular biology and has been home to eight Nobel Prize winners in Physiology or Medicine.
Researchers believe this new AI model could improve the performance of systems in applications like autonomous vehicles, enabling more accurate recognition of pedestrians and other vehicles while significantly reducing energy consumption. The work may also contribute to a better understanding of brain dysfunctions associated with diseases like Alzheimer’s.