Designing Aromas Gets Easier with MIT Algorithms

Companies spending tens of millions of dollars every single year on researching and developing new scents and aromas will soon be able to do so using computer algorithms, rather than humans. The issue here is that the human nose becomes “tired” after a while, which makes it unfeasible for prolonged use.

Big flavor companies are developing scents to be used for a wide variety of applications, ranging from making packaged foods and drinks more appealing to improving the overall appeal of hygiene or cleaning products.

The industry is worth billions annually, and it's easy to understand why using human testers is becoming inconvenient for companies. The nose tends to lose its sensitivity after smelling around 40 samples, which means that tests need to come to a close for that day.

In order to circumvent this obstacle, the large Swiss flavor company Givaudan turned to experts at the Massachusetts Institute of Technology (MIT) Computer Science and Artificial Intelligence Laboratory (CSAIL) for help.

Researchers here used ‘genetic programming’ to crossbreed algorithms in such a manner as to emulate what goes on in the human nose. The main advantage of doing so is that companies will no longer have to struggle to make sense of sometimes-contradictory data supplied by human testers.

“To analyze taste-test results, the CSAIL researchers are using genetic programming, in which mathematical models compete with each other to fit the available data and then cross-pollinate to produce models that are more accurate still,” the MIT team explains.

The research team is made up of Una-May O’Reilly, the principal research scientist at CSAIL, MIT postdoctoral student Kalyan Veeramachaneni and researcher Ekaterina Vladislavleva, who is based at the University of Antwerp.

Details of the new algorithms are published in the latest issue of the scientific journal Genetic Programming and Evolvable Machines. The team admits that it hasn't yet been able to conduct studies aimed at determining whether the new algorithms correctly predict testers’ responses to new flavors.

“People have been playing with these [evolutionary] techniques for decades. One of the reasons that they haven’t made a big splash until recently is that people haven’t really figured out, I think, where they can pay off big,” Hampshire College professor of computer science Lee Spector comments.