Natural-language Semantic Representations for Predicting Human Descriptions of Smell

November 18, 2019
Location: Davis Auditorium, CEPSR
Speaker: Dr. Pablo Meyer Rojas, IBM Thomas J. Watson Research Center


The sense of smell has been ignored and even misbegotten since the positivist revolution of the XIX century.Indeed, studies of comparative neuroanatomy of the famous french neurophysiologist Paul Broca lead him to conjecture that in comparison with that of other animals, the human sense of smell was widely considered weak and underdeveloped. This is based in part on the fact that compared to the rest of the brain the human olfactory bulb is small. A modern look at the human olfactory bulb has the same number of neurons compared to those of rats and mice, which are supposed to possess a superior sense of smell. In fact, the number of olfactory bulb neurons across 24 mammalian species is comparatively similar to that of humans and our sense of smell is similar to that of other mammals. The last 20 years have brought remarkable advances in the science of olfaction, from the discovery of olfactory receptors to the characterization of olfactory circuits, the development of smell detectors or eNoses and our recent breakthrough regarding the predictability of odors from the structure of molecules. Still the prevailing view is that humans’ capacity to identify or characterize odors by name is poor. Paradoxically, I will discuss here that applying natural-language semantic representations allows for the accurate inference, using state-of-the-art machine learning methods and the two largest olfactory psychophysical data sets, of perceptual ratings for mono-molecular odorants over a large and potentially arbitrary set of olfactory descriptors. Combined with a molecule-to-ratings model using chemoinformatic features, our approach allows for the zero-shot learning inference of perceptual ratings for arbitrary molecules.

Speaker Bio

Pablo Meyer RojasPablo is a Team leader in the Translational Systems Biology and Nanobiotechnology group and Research Staff member in the IBM Computational Biology Center. He joined IBM research in 2010 and received his Undergraduate degree in Physics from the University of Mexico UNAM (2000) and a Masters degree from the University of Paris VII/XI his Ph.D. in Genetics from the Rockefeller University (2005). He was awarded a Helen Hay Whitney fellowship as a postdoctoral fellow in Columbia university.


Hosted by Aurel A. Lazar


Subscribe to the CNEC mailing list to get informed about latest news and events.

500 W. 120th St., Mudd Room 524, New York, NY 10027    212-854-5660               
©2014-2019 Columbia University