Large-scale, high-quality human behavioral data is essential for advancing human-centered AI and establishing meaningful benchmarks for the field. To address this need, Hume AI collaborates with researchers and organizations worldwide to co-organize machine learning competitions built on novel, real-world datasets.
Our competitions focus on advancing the understanding of emotional expression and social behavior using multi-modal, multi-task, generative, and few-shot learning approaches. In addition to traditional perception tasks, we increasingly emphasize modeling interpersonal and time-evolving dynamics—such as how affect unfolds over time and across interacting individuals.
Each competition introduces a new dataset capturing underexplored modalities and contexts of human behavior, including voice, facial expression, gesture, self-reported experience, and naturalistic social interactions across diverse, cross-cultural settings.
We welcome participation from both academic and industry teams.
-
See DaiKon2026 for the baseline code of the dyadic interaction challenge at ACII 2026.
-
(Completed) See ExVo2022 for the baseline code provided for the 2022 Expressive Vocalisations Competition held at ICML.
The white paper describing those approaches can be found on arXiv. -
(Completed) See A-VB2022 for the baseline code provided for the 2022 Affective Vocal Bursts Competition held at ACII.
The white paper describing those approaches can be found on arXiv.
More information about our ML competitions can be found on our competitions webpage.