Contributed Talk Sessions | Poster Sessions | All Posters | Search Papers

Poster Session A: Tuesday, August 12, 1:30 – 4:30 pm, de Brug & E‑Hall

Modeling the Hierarchy of the Human Olfactory Perceptual Space via Hyperbolic Embeddings

Aniss Aiman Medbouhi¹, Farzaneh Taleb¹, Giovanni Luca Marchetti¹, Danica Kragic²; ¹KTH Royal Institute of Technology, ²KTH

Presenter: Farzaneh Taleb

Olfactory perception is a complex, high-dimensional process, still largely understudied compared to vision or audition. In this work, we investigate the hierarchical organization of human olfactory perception by embedding perceptual data in hyperbolic space. Hyperbolic geometry, characterized by its exponential volume growth, is particularly well-suited for capturing hierarchical structures. We apply a contrastive learning approach to embed olfactory perceptual data in the Poincaré ball model of hyperbolic space and analyze its structural properties. Our results reveal that odorants with higher perceptual entropy, indicative of greater uncertainty or ambiguity in their perceptual descriptors, tend to be positioned closer to the center of the Poincaré disk, while odorants with lower entropy, reflecting more consistent and distinct perceptual judgments, are mapped toward the boundary. Additionally, individual differences in olfactory perception are reflected in the spatial distribution of embeddings, suggesting that confidence, personality traits, and perceptual biases may influence the way odors are structured in the human olfactory perceptual space. These findings provide a computational framework for modeling olfactory perception. Our approach contributes to the broader goal of understanding the computations underlying sensory perception, bridging cognitive science, neuroscience, and machine learning.

Topic Area: Visual Processing & Computational Vision

Extended Abstract: Full Text PDF