Intel teaches AI smell identification using neuromorphic Loihi chip

Intel uses neuromorphic chip to teach AI to smell.
Image: Intel

On Tuesday, ZDNet reported Intel had a significant breakthrough in training an artificial intelligence (AI) program to replicate a biological function. The chipmaker, in tandem with researchers from Cornell University, designed a platform capable of mimicking the mammalian olfactory system.

In other words, the AI, empowered by Intel’s Loihi neuromorphic processor, can identify different scents the way human beings do.

Intel and Cornell’s Neuromorphic Breakthrough

In 2017, Intel Labs debuted Loihi, a chip designed with neuromorphic architecture. The firm spent six years developing the component, which processes information in the same manner as the brain. Although scientist Carver Mead came up with the notion of neuromorphic computing in the 1980s, the concept’s complexity required decades of technical development before it could be realized.

Manage your supply chain from home with Sourcengine

Indeed, the human brain houses an estimated 100 billion neurons connected via 1 quadrillion synapses. The organ’s biological complexity allows it to process multitudes of data from several inputs very efficiently. If a deep learning program could replicate the brain’s functionality, it would revolutionize computing.

The Loihi processor isn’t that sophisticated yet. It boasts 130,000 synthetic neurons linked by 130 million artificial synapses. But the chip can help AI identify and quantify different scents.

Working together, Intel and Cornell researchers equipped a Loihi-empowered AI program with 72 chemical sensors. The group then pumped 10 different odors, including ammonia and methane, into a wind tunnel to test their system. The scientists found their algorithm could draft neural representations of each scent, even after they introduced “strong background interferents.”

The Potential Applications of Neurotrophic AI

On its face, the advent of machine smelling might not seem like an earth-shattering technological advance. But with further development, Intel theorizes Loihi-based neural networks have applications in medicine, manufacturing, and transportation security.

The chipmaker points out that because certain diseases cause the body to emit specific odors, manufacturers could implement it in next-generation diagnostic tools. The neuromorphic AI can also optimize the function of production facility hazardous material detectors. Besides, developers might embed Loihi-derived chips in transit hub robots designed to identify explosive devices by smell.

Intel believes the neuromorphic technology stack it made with Cornell could be used to develop more robust AI in the future. The team’s breakthrough could help researchers design programs capable of replicating other mammalian senses. The group’s findings might even provide insights that allow computer engineers to create algorithms that address complex abstract problems the way a human being would.

The manufacturer stated that its neuromorphic system wouldn’t be ready for commercial deployment without years of additional research. They still need to teach the algorithm how to detect nuanced olfactory variations, like the different smells of fruit grown in different parts of the world. Nevertheless, the chipmaker’s Loihi processor seems destined for use in a range of groundbreaking new products.


Please enter your comment!
Please enter your name here