Do Computer Chip Really Smells?
What’s Next For Neuromorphic Computing: Senior Intel Research Scientist Nabil Imam tells us all about Neuromorphic Computing.
Nabil Imam from Intel Labs has a hold of a neuromorphic test chip located in his neuromorphic computing lab in Santa Clara, California. Together with his research team based at Cornell University, they are working on building complex mathematical algorithms that can mimic what’s happening in the brain’s neural network while we smell something. (Credited by Walden Kirsch from Intel Corporation)
Senior researcher and scientist Nabil Imam is a member of a neuromorphic computing group from Intel Labs. At Cornell University, he is working together with top neurophysiologists who specialize in olfaction. Imam, who has a doctorate in neuromorphic computing, tells us that he and his friends from Cornell study the olfactory system’s biology by running tests on animals.
They measure all electrical activity happening in their brains while introducing them to different smell odor. Based on the results from these electrical pulses and circuit diagrams, they derive algorithms and configure them around neuromorphic silicon, much like the Loihi test chip. The Loihi is basically Intel’s neuromorphic chip. It can apply all these computation principles located in our biological brain and adapt them to computer architectures.
Just recently, Nature Machine Intelligence made a research profile led by Intel and Cornell’s scientists focusing on the way they build mathematical algorithms. Thanks to the guidance of the researchers, Loihi quickly managed to learn the neural representations of multiple different smells and odor.
How do we smell?:
Let’s say you picked up a grapefruit and then took a whiff. What happens next is that the molecules in the fruit stimulate your nose’s olfactory cells(The word olfactory has a Latin origin, and it means to smell something). The cells located in your nose then send out signals to your brain. After that, the olfactory system registers these electrical pulses, and a corresponding link of neurons create what we call a smell sensation.
No matter what you’re smelling, be it toxic fumes or roses, a system of neurons in your brain formulate specific sensations that are related to the objects. Identically, your sense of sound and sight, revocation of memories, decisions, and emotions each have corresponding neural networks that you compute in different ways.
Loihi is continuously learning to detect distinctive odors in complex blends: Imam and his research team looked at a data collection consisting of 72 chemical sensors. More specifically, they looked at the activity of these sensors when exposed to 10 noxious odors inside a wind tunnel.
They later submitted the sensors’ response for each particular scent to Loihi, where circuits made out of silicon replicated the brain’s circuitry that gives us the smell sense. The chip quickly learned the neural representation of each of the smells, including methane, ammonia, and acetone, and managed to identify them regardless of the strong interferences running in the background.
Carbon and smoke detectors in your home use similar sensors for the detection of these odors but cannot distinguish them; They can detect harmful substances in the air but cannot intelligently categorize them.
According to Imam, the community of chemical-sensing has long been searching for quick, reliable, and efficient chemosensory processing systems, also known as “electronic nose systems.” He recognises machines’ potential when attired with neuromorphic chips to monitor the environment, detect hazardous materials, or do simple quality control tasks in factories.
Get more technology updates@Techology
Alternatively, they can make medical diagnoses based on the fact that some diseases can emit specific odors or even identify dangerous substances at airports.
Adding more senses down the line: Imam tells us that his next step is to apply this same approach to a broader range of problems. For example, sensory scene analysis (trying to understand the relation between observable objects) or more complex or abstract issues like decisions and plans.
Imam also thinks that once we understand how neural circuits in our brains can solve difficult computing problems, we will have all the clues necessary to design an efficient and powerful machine intelligence.
Difficulties to overcome:
There are still many difficulties related to olfactory sensing, according to Imam. When walking into a grocery store, you might stumble upon the smell of strawberries.
The problem is that this smell is much like the one you might sense from a banana or blueberry; neural activity patterns formulated in our brains are very similar to each other in such situations. It’s hard for humans to make out a difference between a fruit and a mixture of numerous scents.
Systems can get confused when trying to smell fruits and then put them in the same category. Smelling strawberries from Italy and then from California will prove challenging to label together in the same class because they’ll have different corresponding aromas.
Imam hopes that these challenges will be solvable in the coming years. He considers his work a fine example of breakthrough research taking place in the sphere between artificial intelligence and neuroscience.