March 17, 2020 – A paper jointly published by Intel Research and Cornell University researchers in the journal “Nature Machine Intelligence” shows that Intel’s neuromorphic research chip Loihi Ability to learn and identify hazardous chemicals in the presence of significant noise and obscurity. Loihi can learn to recognize each smell with just a single sample, without destroying its memory of previously learned smells. Compared with traditional state-of-the-art methods, Loihi exhibits extremely good recognition accuracy. A deep learning solution is included in the traditional approach, but to achieve the same classification accuracy as Loihi, the solution requires 3,000 times more training samples to learn each odor.
Nabil Imam, senior research scientist at Intel’s Neuromorphic Computing Lab, said: “We are developing neural algorithms at Loihi to simulate how the human brain works when we smell. This work represents contemporary research at the intersection of neuroscience and artificial intelligence. An example, and confirms that Loihi has the potential to provide important perception capabilities and make them beneficial to all walks of life.”
This close-up photo shows Loihi, Intel’s neuromorphic research chip. Intel’s latest neuromorphic system, Pohoiki Beach, will consist of 64 Loihi chips. Pohoiki Beach has been launched in July 2019. (Source: Tim Herman/Intel Corporation)
Study content: Intel and Cornell University researchers used a set of neural algorithms derived from the structure and dynamics of the brain’s olfactory circuit to train the Intel Loihi neuromorphic chip to learn and recognize 10 hazardous chemicals smell. To do this, the research team took a dataset of 72 chemical sensor activities in response to these odors, and deployed a circuit diagram of biological olfaction on the Loihi chip. The chip quickly mastered the neural representation of each odor, and could identify each odor even under obvious masking conditions, which fully demonstrated the broad prospects of future research at the intersection of neuroscience and artificial intelligence.
For more background information see: Nature-Machine Intelligence | Intel Neuromorphic Computing | Intel Research
The Links: 7MBP50RA060-06 VI-2T0-EW