The Project

Despite the advances in Affective Computing, the application of affect models is still narrowly limited to answering questions like: "What is the mental state of patients while interacting with a robot?", "What emotions does an advertisement elicit in viewers?", "What is an actor's emotion in a scene?" or "What is the user's emotional state while playing a game?". From the above, it becomes apparent that affect models are context-dependent and cannot answer questions like "Why do these emotional states appear?" or "What are the elements of the context that cause the specific emotions?". The next generation of affect models should be context-invariant and able to identify and exploit cause-effect relations in humans’ emotional manifestation. ERICA is an ambitious project that aims to design affect models that use the ubiquitous cause-effect mechanisms in humans’ emotional expression.

The main challenges that hinder the development of causation-aware affect modelling algorithms are:

• - the lack of benchmarking datasets for causal inference,
• - the presence of causal and anti-causal relations between stimuli and emotions,
• - the lack of standard methodologies for learning causal representations from low-level data commonly used in affect modelling, and
• - the fact that causal affect models should be modular, which is in stark contrast to current state-of-the-art approaches that propose huge monolithic models.

ERICA proposes a holistic approach for causal affect modelling ranging from the identification of high-level causal predictors in affect modelling, up to developing modular models of affect based on the identified cause-effect mechanisms.

ERICA addresses fundamental research questions in affect modelling. Its results will serve for ground-breaking technological advances in affective computing and emotion-aware artificial intelligence.