Emotion Recognition with Machine Learning

My MS students work closely with stellar PhDs and PostDocs and world-class international collaborators. Some examples:  

 

Røskva Bjørgfinsdóttir and Brage Eilertsen (first authors of AAAI submission over a summer project). first job after graduation: somewhere very decent! 
 

Iman Tabrizian (co-first author of NeurIPS paper). first job after graduation: full-time engineer at NVIDIA   

 

Wanyun Xie (first author of TMLR paper). first job after graduation: Ph.D. at EPFL

 

 

Human emotion, reflected as gut feeling in the body and the way we interpret and appraise a situation in our mind can be defined as complex patterns of responses to events that matter to us. Human emotions has three types of joint and interactive elements: 1) Physiological elements via body heart rate, breathing, and hormone release; 2) Behavioral elements via facial expressions and actions in response to a stimulus; 3) Cognitive elements via the way we appraise and interpret a thing or situation. Different from long-lasting mood and general term affect, “emotion” is an immediate, brief, and specific response to an internal or external stimulus.
 

Multi-modal sensory data have been used to classify arousal, valence, and emotion using e.g., contrastive language signal pre-training. 
 

We extract human emotions obtained from multiple modalities and physiological, behavioral, and cognitive elements considering both “Discrete Emotion Theory” and “Core Affect Circumplex Model”. 
 

[1] P. Singh, R. Budhiraja, A. Gupta, A. Goswami, M. Kumar, & P. Singh, “EEVR: A dataset of paired physiological signals and textual descriptions for joint emotion representation learning,” NeurIPS, 2024.

[2] D. L. Schacter, D. T. Gilbert, & D. M. Wegner, Psychology. Macmillan, 2009.

[3] P. Ekman, “Emotions revealed,” Bmj, 2004.

[4] J. A. Russell, “A circumplex model of affect.” Journal of personality and social psychology, 1980.

 

Published Nov. 2, 2025 7:38 AM - Last modified Nov. 2, 2025 7:38 AM

Supervisor(s)

Scope (credits)

60