Machine learning: An algorithm designed to smell

A neural algorithm designed to mimic the biological perception of smell is reported in a paper published in Nature Machine Intelligence. These findings may support a future application in which an artificial nose can be trained to recognize specific odours, despite the presence of unknown background odours.

Neuromorphic chips are designed to use computational machinery inspired by the brain, namely by creating networks that consist of artificial neurons and synapses. However, it is unclear how to use that machinery in real-world practical problems. This is largely due to our incomplete understanding of algorithms implemented at the level of biological neural circuits.

Nabil Imam and Thomas Cleland describe a neural algorithm for the learning and identification of odour samples based on the architecture of the mammalian olfactory system. The neural algorithm was then implemented into a neuromorphic system, where it was trained on scents - such as toluene, ammonia, acetone, carbon monoxide and methane - and tested on data from sensors in a wind tunnel.

These results reveal computational features for understanding mammalian olfaction, as well as for improving the performance of artificial chemosensory systems. The findings also suggest that the adaptation of such biological neural systems could represent a powerful method to develop new algorithms that go beyond current trends in artificial intelligence.

doi:10.1038/s42256-020-0159-4

Return to research highlights