Japanese scientists develop AI that can automatically generate scents
Fragrance is an important factor in many products, including beauty products, food, and everyday items. However, creating a new fragrance can take time and sometimes requires the help of a professional perfumer. Research at Tokyo University of Science has created OGDiffusion , an AI that takes over this time-consuming and labor-intensive process and automatically generates new fragrances.
Generative Diffusion Network for Creating Scents | IEEE Journals & Magazine | IEEE Xplore
Generative AI masters the art of scent creation | Science Tokyo
'Odor (Aroma) Generative Diffusion (OGDiffusion),' developed by Takamichi Nakamoto and others at Tokyo University of Science, is an AI model that calculates the essential oil combination needed to reproduce a desired scent when the user specifies the characteristics of that scent.
Nakamoto and his team used 166 types of essential oils and assigned nine types of 'aroma descriptors' to each scent, such as 'herbal,' 'floral,' 'woody,' and 'sweet.' By training OGDiffusion with these scent descriptors and the mass spectrometry data of the essential oils, they were able to successfully reproduce the scents.
In experiments where random noise was added to the mass spectrometry data and OGDiffusion was then instructed to calculate the original mass spectrometry data using the mass spectrometry data and scent descriptors, we were able to successfully recover the original mass spectrometry data in a high percentage of cases.
Next, OGDiffusion was used to generate scents with multiple scent descriptors, which were then reproduced and smelled by 14 subjects. The researchers then found that a significantly higher percentage of subjects identified the scent descriptors as the same as the instructions given to OGDiffusion.
Furthermore, in an experiment in which subjects were asked to choose between a scent that reproduced a specific scent descriptor and a scent that did not contain the specific scent descriptor and to select which matched the specific scent descriptor, the subjects consistently made the correct choice. These experiments show that OGDiffusion can reproduce the desired scent descriptor with high accuracy.
There are other models that generate scents using AI besides OGDiffusion, but they rely on proprietary data sets and still require experts to use AI. On the other hand, OGDiffusion is characterized by its ability to easily generate new scents, and since it generates scents based on essential oil recipes, it is also easy to reproduce scents.
Nakamoto said, 'OGDiffusion automates scent generation, providing an efficient way to create scents. Furthermore, even non-experts can create the intended scent. OGDiffusion is the first scent generation model of its kind, and it heralds a future in which AI will change scent design.'
Related Posts:
in Science, Posted by log1p_kr