Abstract:Public olfaction datasets are small and fragmented across single molecules and mixtures, limiting learning of generalizable odor representations. Recent works either learn single-molecule embeddings or address mixtures via similarity or pairwise label prediction, leaving representations separate and unaligned. In this work, we propose AROMMA, a framework that learns a unified embedding space for single molecules and two-molecule mixtures. Each molecule is encoded by a chemical foundation model and the mixtures are composed by an attention-based aggregator, ensuring both permutation invariance and asymmetric molecular interactions. We further align odor descriptor sets using knowledge distillation and class-aware pseudo-labeling to enrich missing mixture annotations. AROMMA achieves state-of-the-art performance in both single-molecule and molecule-pair datasets, with up to 19.1% AUROC improvement, demonstrating a robust generalization in two domains.
Abstract:To enhance the robustness and resilience of wireless communication and meet performance requirements, various environment-reflecting metrics, such as the signal-to-noise ratio (SNR), are utilized as the system parameter. To obtain these metrics, training signals such as pilot sequences are generally employed. However, the rapid fluctuations of the millimeter-wave (mmWave) propagation channel often degrade the accuracy of such estimations. To address this challenge, various blind estimators that operate without pilot have been considered as potential solutions. However, these algorithms often involve a training phase for machine learning or a large number of iterations, which implies prohibitive computational complexity, making them difficult to employ for real-time services and the system less resilient to dynamic environment variation. In this paper, we propose blind estimators for average noise power, signal power, SNR, and mean-square error (MSE) that do not require knowledge of the ground-truth signal or involve high computational complexity. The proposed algorithm leverages the inherent sparsity of mmWave channel in beamspace domain, which makes the signal and noise power components more distinguishable.