The most stringent annotation protocol
A cornerstone of good AI is the quality of the data used to train the model.
Vima’s algorithms, which assess traits, soft skills, and emotions, meet the best standards of data quality. In effect, they are built on the human annotation of videos according to a scientifically-validated protocol and a careful selection of a large group of expert annotators (psychology background, half male/female, verified annotation reliability).
Vima reduces biased assessment by measuring multimodal expressions of traits, states, and soft skills through behaviors. We indeed focus on what you do, not what you say you do.
In addition, Vima works relentlessly to train its algorithms on quality data both in terms of video quality and human labelling, upholding the principle of “quality in, quality out”. For example, individual results are compared to a norm group with similar language and geographical background.
Prediction engine validity
Designing a cross-disciplinary system, such as Vima’s behavioral analysis engine, is crucial to access the validity of both its training data (annotation/ratings of the behavioral traits and skills) and the predictive performance of the trained Machine Learning system.