Computer Says- You Smell

Can you imagine your computer asking you to go and take a shower? Or instead, maybe your laptop will gauge the odds of your new cologne helping you meet ladies at the bar. While this may sound like science-fiction, scientists at the Weizmann Institute in Israel have succeeded in training the first electronic system that can predict the pleasantness of new odors.

In research published in PLoS Computational Biology, scientists argue that the perception of an odor’s pleasantness is innately hard-wired to its molecular structure, with other factors like culture assisting in the final fine tuning process.

So how does this eNose work?  The main component of an eNose is an array of chemical sensors. As an odor passes through the eNose, its molecular features stimulate the sensors in such a way as to produce a unique electrical pattern – an ‘odor fingerprint’ – that characterizes that specific odor. Like a dog sniffing around for explosives or drugs, an eNose first needs to be trained with odor samples so as to build a database of reference. Then the instrument can recognize new samples of those odors by comparing the odor’s fingerprint to those contained in its database.

Till now the only back draw to the eNose was the fact that it could only recognize smells that were introduced to it beforehand. To solve this problem, a team of Weizmann scientists, led by Dr. Rafi Haddad, Prof. Noam Sobel of the Neurobiology Department, and Prof. David Harel of the Computer Science and Applied Mathematics Department, decided to approach this issue from a different perspective. Rather than train an eNose to recognize a particular odor, they trained it to estimate the odor along a particular perceptual axis allowing for the first “free thinking” eNose.

To achieve this, the scientists first asked a group of native Israelis to rate the pleasantness of a selection of odors according to a 30-point scale ranging from “very pleasant” to “very unpleasant.” From this dataset, they developed an “odor pleasantness” algorithm, which they then programmed into the eNose. The scientists then got the eNose to predict the pleasantness of a completely new set of odors not contained in their database against the ratings provided by a completely different group of native Israelis. The scientists found that the eNose was able to generalize and rate the pleasantness of odors it never smelled before, and these ratings were about 80% similar to ratings given by people who had not participated in the eNose training phase. Moreover, if the odors were simply categorized as either “pleasant’” or “unpleasant,” as opposed to being rated on a scale, it achieved an accuracy of 99%.

To try and avoid a cultural skew the scientists decided to test eNose predictions against a group of recent immigrants to Israel from Ethiopia. The results showed that the eNose’s ability to predict the pleasantness of novel odors against the native Ethiopians’ ratings was just as good, even though it was “tuned” to the pleasantness of odors as perceived by native Israelis. “Being able to predict whether a person who we never tested before would like a specific odorant, no matter their cultural background, provides evidence that odor pleasantness is a fundamental biological property, and that certain aspects of molecular structure are what determine whether an odor is pleasant or not. For instance, many may wonder how the French can like the smell of their cheese, when most find the smell quite repulsive. We believe that it is not that the French think the smell is pleasant per se, they merely think it is a sign of good cheese. However, if the smell was presented out of context in a jar, then the French would probably rate the odor just as unpleasant as anyone else would.”

These findings have important implications for automated environmental toxicity and malodor monitoring, fast odor screening in the perfume industry, and mark a crucial step towards the Holy Grail of sense technology – transmitting scent digitally.

1 person likes this post.

Unlike