Deep learning allows Affectiva to model more complex problems with higher accuracy than other machine learning techniques. It solves a variety of problems such as classification, segmentation and temporal modeling. It also allows for end-to-end learning of one or more complex tasks jointly including: face detection and tracking, speaker diarization, voice-activity detection, and emotion classification from face and voice.
Affectiva has already analyzed over 7.5 million faces in 87 countries and has used this data to build its core Emotion AI technology. We also collect large amounts of spontaneous driver and occupant data so we can tune our existing algorithms for automotive environments, and develop new automotive metrics. We are building this large automotive data corpus through in-car data collection, lab simulation environments and data partnerships.
Our deep learning models provide accurate, real time estimates of emotions and cognitive states on mobile devices and embedded systems. Our focus is on building highly efficient models that achieve high accuracy with a minimal footprint. Our approach involves joint-training across a variety of tasks using shared layers between models combined with iterative benchmarking / profiling of on-device performance, and model compression - training compact models from larger models.
