Computational Research

Perfumery has a long history of great structures. The chypre, the fougère and the amber each carry more than a century of refinement, and most modern fragrances descend from a small number of them. My research asks a simple question: what lies between them?

Exploration at scale

I simulate millions of fragrance formulas that have never existed and analyze how they relate to existing structures. This large-scale exploration reveals new compositions that lie between classical forms.

Materials as data

Every raw material has measurable properties: how quickly it evaporates, how strongly it is perceived, and how it behaves alongside other materials. I maintain a detailed record of these properties, so a formula can be judged on how it will unfold over hours, not only on how it smells in the first minute.

Seeing a fragrance unfold

A fragrance is not a fixed object. Its character shifts as each material rises, peaks and fades. I developed cloud plots to show that change, mapping how a composition's perceived character moves from the first spray to the final drydown.

Visualizing How a Fragrance Evolves Perceptually: Cloud Plots

Structure and lineage

Studying thousands of established compositions shows which materials anchor a family and which only decorate it. That map lets a new formula be placed exactly: close to a classic, or deliberately away from it.

The hand decides

To identify top candidates, I select the most promising structures, blend them, and scent-test them by hand. Computation widens what can be considered. The final judgment is always made on skin and blotter.

Background

For more than three decades I have written scientific algorithms from first principles, and for over twenty years that has included neural networks, machine learning, Monte Carlo simulation and text mining. Every method I use is my own code; I don't use third-party libraries or packaged tools. My earlier work applied them to radiation dosimetry for NASA astronauts, cancer genomics and clinical trials, across faculty appointments at Baylor College of Medicine, Weill Cornell Medical College and Houston Methodist Research Institute. I served as Editor-in-Chief of Source Code for Biology and Medicine from 2006 to 2020, and I have published more than 100 peer-reviewed articles.

Nodus applies the same discipline to fragrance. Every formula the software proposes is blended and evaluated by hand.

Professional affiliations

• Senior Member, Institute of Electrical and Electronics Engineers (IEEE)
• Senior Member, IEEE Computational Intelligence Society
• Senior Member, Società Italiana Reti Neuroniche (SIREN), the Italian Neural Network Society
• Member, International Neural Network Society (INNS)