AI・香りの新技術

Making scent arrive as you approach a virtual flower takes more than a fast device. This study combines miniature scent generators with motion prediction to account for scent travel time.

Delivering scents in time with video does not ensure that viewers interpret them as intended. A four-scent display highlights both delivery design and the challenge of making coffee smell like coffee.

A Principal Odor Map predicts odor descriptions from molecular structure. Human testing reveals its strengths and limits for unfamiliar molecules.

An AI used essential-oil data to search for scent formulas. Human tests supported adding floral character, but success varied across odor features.

A 2026 study predicted odor strength from molecular structure. AI trained on 2,393 molecules in four categories, odorless, low, medium, and high, achieved 72% agreement in testing. We examine its predictive clues and limitations concerning concentration and perfume applications.

Can the NMR fingerprint of perfume ingredients help identify authentic products? A study of 28 samples illustrates how to read the signals and why variation among authentic products matters.

One-pot Bayesian optimization reduces sample use while matching sensor responses. Matching a machine signal does not guarantee the same perceived odor.

MultiScent-20 showed preliminary validity and repeatability for odor identification, with limits in population coverage and clinical interpretation.

Pyrfume connects olfactory datasets while preserving stimulus conditions and measurement meaning, providing a foundation for research and AI.

A light-based technique measures chirality in perfume vapour and follows changes in pine emissions, without ranking fragrance quality or proving authenticity.