Researchers at Intel Labs and Cornell University have just demonstrated the unique capabilities of Loihi, an Intel neuromorphic research chip, to identify a variety of harmful chemicals by smell alone. Loihi analyzes and identifies every chemical in the test sample without destroying previously learned memories of odors, the researchers said. Loihi also shows greater accuracy than traditional identification systems, including deep learning.
(Pictured: Intel Labs)
In contrast, deep learning systems require approximately 3000 times the sample training to reach the same level as Loihi. Nabil Imam, senior research scientist at Intel Labs, says:
We are developing neural algorithms on Loihi to mimic the associated reactions that occur in the brain after smelling.
This work is a model of the intersection of contemporary neuroscience and artificial intelligence, demonstrating Loihi’s potential to deliver important sensing capabilities that benefit all industries.
It is reported that as a hardware, the Intel Loihi chip is designed to mimic how the human brain handles and solves problems. It was first announced in September 2017, when Intel called it an “incredible learning speed.”
The chip is unique in its ability to infer new data using known knowledge, which accelerates its learning process exponentially over time.
The Loihi chip, designed with a architecture based on ‘neuromorphic computing’, was inspired by scientists’ understanding of the human brain and its latest problem solving research.
According to a study published today in the journal Nature Machine Intelligence, it can be seen how the Intel lab and Cornell University’s research team built the neural algorithm skewed from scratch based on the structure and dynamics of the human brain’s olfactory circuit.
The chip now learns and identifies the odors of 10 different dangerous chemicals, the same principle behind which the human brain perceives different odors.
For example, when a person picks up a grapefruit and smells it, the fruit’s molecule stimulates the olfactory cells in the nose and sends the relevant signals to the brain.
Electrical impulses in interconnected neurons can then produce a unique feeling about the smell.
“Whether you smell grapefruit, rose, or harmful gas, the network of neurons in the brain produces a feeling specific to the object,” the Intel researchers explained.
Visual and auditory feelings are similar, with the brain’s memory, interests, and decisions all being associated with neural networks and calculations in specific ways.
In the latest study, the Intel team used a data set that included the activity of 72 known chemical receptors in the brain and how they responded to the chemical odors of each substance.
The team used the data for so-called ‘biool circuits’ on Loihi to enable Loihi to identify the neural circuits of each odor.
Patrick Moorhead, an analyst at Moor Insights and Strategy, told Silicon ANGLE that the study is a great example of determining the neural morphology of the odors of various harmful chemicals.
Looking ahead, the technology could be used as an ‘electronic nasal system’ to help doctors diagnose a variety of diseases. Other uses include the development of more efficient smoke/carbon monoxide alarms, or explosive biodetecting systems at airports.
Next, the team also wants to extend the technology to more problem solutions, from sensory scenario analysis (understanding the connections between observed objects) to abstract problems (such as planning and decision-making).