Plume Tracking and Odor Mapping Algorithms

A deep reinforcement learning model that allows AI agents to track odor plumes

Feb 2023, phys.org

https://techxplore.com/news/2023-02-deep-ai-agents-track-odor.html

Insects track odor plumes to find mates. (And there was a similar study done recently here.)

"Instead of running a traditional laboratory wind-tunnel experiment, we used a complementary 'in-silico' approach using artificial neural networks," Singh explained. "This helped us develop an integrative understanding of plume tracking across multiple levels, including emergent behavior, neural representation and neural dynamics."

To train their plume-tracking agents using DRL, the researchers first simulated an odor emanating from a source located within a windy arena with a total area of approximately 120 m2. When their agents identified where the source of the odor was located, they received a reward. In contrast, if they lost track of the odor plume and left the arena, they were "punished."

"The behavior that emerges in our trained artificial agents bears a striking resemblance to the behavior modules biologists have previously observed in flying insects performing plume tracking," Singh said.

via University of Washington and University of Nevada: Satpreet H. Singh et al, Emergent behaviour and neural dynamics in artificial agents tracking odour plumes, Nature Machine Intelligence (2023). DOI: 10.1038/s42256-022-00599-w

https://dx.doi.org/10.1038/s42256-022-00599-w

Image credit: AI Art - Total Hallucination - 2023

Revealing Salient Features of the Smell Network

aka On Sweetness and Flowers

pdf screenshot: Aldrich's Flavors and Fragrances Catalog, 2013. "Organoleptic properties index", pp145-169. 

If you work in the Flavor and Fragrance Industry, chances are you use the Sigma Aldrich catalog to reference the chemicals, the raw materials, of your craft. Sigma Aldrich produces and procures about 1,000 ‘smellable’ chemicals for distribution among the Industry.

It is the nature of volatile organic compounds (‘smellable’ molecules) that the same units can be used to make a variety of aromas. Although there are only 1,000 chemicals, they are indexed by over 3,000 descriptors, such as smoky, anise, spicy, or rose. One does not search for the chemical Geraniol, but instead looks up “Rose”, under the super-category Floral, and there will be Geraniol, along with all other chemicals used to intimate the aromatic impression of a rose. Further collapsing these roughly 3,000 descriptors into more general categories, again there is redundancy. The aromatic descriptor “Lemon” is categorized under the Citrus as well as the Woody groups.

Of all these descriptors, “Sweet” is the most redundant; it is repeated in four different groups, more than any other descriptor. In network science parlance, this is called centrality, or betweenness. Thus, Sweet has a higher centrality value than any other descriptor. It is a kind of glue, a hub, which holds the network together. (This is a flavor and fragrance catalog, after all.)

Using Sweet as a locus, a modified aroma network can be configured. Within this “Sweet network”, every super-category that contains Sweet as a descriptor is present. (Note that the chemical indexed by “Sweet” in one grouping, such as Fruity, does not necessarily refer to the same chemical as the “Sweet” in the Balsamic group.) The related super-categories that share Sweet as one of their descriptors, they happen to be the most ‘copiously described’ groups. Fruity has 850 descriptors, the most of any group. Balsamic has 540.

This revelation is becoming somewhat anti-climatic, as it would be expected that the more descriptors a group has, the more likely it would share one of those descriptors with another group. But what is this? If we look down, at the third-most populated grouping – Floral, at 296 – we see that it is absent from the Sweet network. If groupings with many descriptors tend to contain Sweet, and Floral has many descriptors, then it should also contain Sweet. It does not.

In the first place, Sweet is not a smell, it is a taste. But this is a flavor and fragrance catalog, you retort. But what then about sour, salty, bitter, and savory? They do not exhibit the same centrality as sweetness. Next, we do not eat flowers. In fact, a major part of the profile of a “floral” note is the smell of shit – not the most gustatorily satisfying of aromas.

The corpus of the Sigma Aldrich smell network is highly complex in its structure and associations, with stable patterns found few and far between. No matter the corpus, throughout history such has been the same for smells. At least for one of its infinite configurations, it can be observed that Sweetness and Flowers are very distinct entities in the aromasphere: We do not eat Flowers, and we do not smell Sweet.

Age of Approximation

Screenshot from Iain McGilchrist called the Divided Brain on RSA Animate and TED.

“The Age of Enlightenment is Dead.” Thank you, Mr. Danny Hillis, for putting this in writing, and in the new MIT Journal of Design and Science no less.

Mr. Hillis makes the case, in a brief but very coherent treatise, that science is due for an update. In fact, it is not just science, but the very idea of human endeavor and progress. I recall the TED talk given by Iain McGilchrist called the Divided Brain. (This guy is author of The Master and His Emissary: The Divided Brain and the Making of the Western World, 2009). In his talk, he describes how the brain is split in two; the truth is more nuanced than that, and this is the point of the talk in fact. He goes on – there are two metaphorical sides, and they each do two different things, and the reason we tend to be right-handed is because the corresponding side of our brain is for active, intentional manipulation. And on – throughout history the pendulum swings, and one day we may regard the right side as the “right” side (because the Age of Enlightenment made us value the left side so much).

In the art classroom, I repeat these ideas, and ask them what the world would be like if everyone was “right-brained,” or artistically minded. Imagine if everyone was an artist, and nobody a scientist, all emotion and flight-of-fancy, and no bridges, tunnels, infrastructure, economic policy, institutions of higher learning, no numbers, no logic, and nothing to separate sense from nonsense. Crazy.

Then again, I can sort of imagine a world where artificial intelligence does all that stuff for us, so we can do more human things, more messy, emotional, intuitive things. I highly doubt this is what Mr. Hillis is talking about in his essay, but I’m quite excited nonetheless that he’s talking, period. He says right here:

“As our technological and institutional creations have become more complex, our relationship to them has changed. We now relate to them as we once related to nature. Instead of being masters of our creations, we have learned to bargain with them, cajoling and guiding them in the general direction of our goals. We have built our own jungle, and it has a life of its own.”

The Enlightenment is Dead, Long Live the Entanglement

Danny Hillis, MIT Journal of Design and Science, March 2016

Post Script:

What the hell does all this have to do with the Language of Smell?

A primary objective of Hidden Scents is to present the idea that after the wave of Big Data crashes on the shores of human civilization we will have entered a new era, one in which certainty itself is no longer valued in the way it once was. We already see this today, when we ask what it is that separates us from our imminent AI overlords. Humans have intuition, something an algorithm can never have, by its nature. Humans can do this thing called “messy thinking,” or fuzzy thinking, or half-thinking. This is what leads us to make novel discoveries and connections and to be creative in general. This is what makes us not computers. And in its uncanny way, this kind of mental activity is at the core of olfaction. To smell something is to navigate a sea of data too large to fully comprehend. In this sea, one can approximate, but never ascertain. (The source of a particular smell is only verified by one of the other senses, like when you actually find that dead mouse under the fridge.)

So if the Age of the Enlightenment is dead, then perhaps olfaction (and more specifically the language of olfaction) can serve to carve the path ahead.

Well, perhaps you think I’m a bit too right-brained to be writing about such things. Thanks for reading at least.