How Do You Teach A Computer To Smell?
March 25, 2020Our sense of smel is one of five basic senses gifted to most of us. Something as trivial as picking up a grapefruit and identifying it’s scent, actually involves a complex network of biological systems within our nose and brain.
When you smell said grapefruit, the fruit’s molecules stimulate olfactory cells in your nose (the word olfactory is derived from the Latin word olfactare, which means “to smell”). These cells, then send signals to the brain’s olfactory system.
Here, electrical pulses within an interconnected group of neurons generate a smell sensation. This is the same mechanism that that is used to identify specific odors and allow the brain to associate that smell with the respective object.
Intel’s researchers are using similar principles to create a computer with the ability to smell. Nabil Imam a senior research scientist in Intel Labs’ neuromorphic computing group, works with olfactory neurophysiologists at Cornell University.
The team at the university study the biological olfactory system in animals, then measure the electrical activity of the brains as they smell odors. The data is then used to identify the circuit diagrams and electrical impulses. Imam and his team at Intel uses this data to derive a set of algorithms.
They then configured them on a neuromorphic silicon known as the Loihi test chip. This is Intel’s neuromorphic computing chip, which is a computered adaption of the human brain. The chip maker is training it to identify 10 different odors.
Imam and team collected data from 72 chemical sensors that were exposed to 10 gaseous odors circulating within a wind tunnel. The sensor readings of each odor was transmitted to Loihi, which then analyzed the information to successfully identify each one, despite the presence of strong background interferents.
Devices like smoke and carbon monoxide detectors use sensors to identify odors but they aren’t able to distinguish between them. Loihi, is the next evolutionary leap forward in terms of odor detection technology. And with it, robots could detect hazardous materials, aid medical diagnosis and solve other real world problems.