Working on the emotions of people, Marianne Reddan was involved in the process of producing a computer that could detect people’s emotions. Reddan devoted his years to this issue, how the machine was designed and how it came to the final transfer.
Marianne Reddan has devoted her last decade to human faces. Reddan, who studied two very close feelings of surprise and fear, especially after working on these two emotions, can no longer distinguish between two emotions.
Reddan, therefore, developed a system at Stanford University with the help of his friend who was interested in neuroscience. With this system that uses machine learning, a distinction has been made between the two emotions.
(Left fear, right astonishment)
When Reddan examined the system, he shared a point that made him extremely surprised. The EmoNet system not only looked at people’s faces in order to understand how people experience emotions but also included environmental factors as expected from a person.
“If EmoNet can distinguish the feeling of surprise and fear,” Reddan said in an interview with The Daily Beast, he successfully analyzed not only the differences between faces but also the differences between environments.
In order to establish a machine learning system that can distinguish people’s emotions, previously established data sets were received and the system took a year to develop. Reddan and colleagues initially proposed to detect emotions using a deep learning model called AlexNet that detects objects on computers.
Reddan, who sees the idea of learning emotions as a challenging task and wants to try it, decided to train this model by choosing this way. The model was taught about 20 emotions.
Although common feelings such as ‘anxiety’ or ‘boredom’ were included in these feelings, rare emotions such as ‘aesthetic appreciation’ and ’empathy’ were also taught. Neural networks, which examine all the visuals, did research from people’s facial expressions to body language and put emotions into place.
According to the results of the study, the model was brought in front of an MRI machine. The MRI machine performed a mapping based on the brain activity of real persons. 112 visuals were shown to the subjects and they were determined which emotion they felt.
The model, which successfully passed this test and took the name EmoNet, became one of the few models that categorized different emotions and gave reliable results.