ACLab works at the intersection of computer vision and machine learning. We are interested in representation learning algorithms for visual perception (object recognition, localization, segmentation, pose estimation, activity tracking, ...) with the multidisciplinary goal of understanding, detecting, and predicting human behaviors by estimating their physical, physiological and emotional states. For a robust and efficient state estimation, we represent the state of the world in a low-dimensional embedding, called "pose", which is a succinct interpretable representation of the important information in the state. At ACLab, we use machine intelligence (mainly Computer Vision and Machine Learning) to solve these pose estimation problems and to give human leverage, not to replace them! At ACLab, we are also working on problems in Small Data domains. To deal with data limitation, we do integrate explicit (structural or data-driven) domain knowledge into the learning process via generative models, while benefiting from the recent advancements in data efficient ML.