2025-04-23 東京科学大学
＜関連情報＞
- https://www.isct.ac.jp/ja/news/zj4km15ao3vi#top
- https://www.isct.ac.jp/plugins/cms/component_download_file.php?type=2&pageId=&contentsId=1&contentsDataId=1407&prevId=&key=04c2e0b804df44800294f7036a1b3d61.pdf
- https://ieeexplore.ieee.org/document/10943150
香りを創り出す生成拡散ネットワーク Generative Diffusion Network for Creating Scents
Manuel Aleixandre; Dani Prasetyawan; Takamichi Nakamoto
IEEE Access Published:27 March 2025
DOI:https://doi.org/10.1109/ACCESS.2025.3555273
OGDiffusion generates new scents by conditioning mass spectrometry (MS) profiles on odor descriptors. Using Pine Scotch essential oil as a reference (green circle), the network reproduces its MS profile (blue squares) and creates new spectra with modified odor descriptors (orange and yellow stars). Principal Component Analysis (PCA) shows how changes in descriptors shift the generated MS profiles, enabling targeted fragrance design.
Abstract
This paper introduces a novel application of generative diffusion networks for creating scents. The research presents a generative diffusion network designed to create new aromas with specified, required odor descriptors using essential oils as the basic components. The model uses mass spectrometry data of essential oils as latent embedding space of the essential oils. The generative network outputs mass spectrometry data as the primary output. These generated mass spectrometry profiles are then processed by non-negative least squares to create essential oils recipes that have the required odor descriptors. The results demonstrate the model’s ability to produce diverse and new aroma profiles, which are validated by sensory tests. The method can create new scents by mixing essential oils, making automated aroma design possible. This approach shows major progress in aroma design. These results suggest many uses in industries like perfumery and food and beverage, improving efficiency and creativity in making many different fragrances.
