Psychotherapist Baev explains why neural networks “don’t understand” facial human emotions

No time to read?
Get a summary

Neural networks are useless for analyzing human behavior as they cannot accurately recognize emotions. In addition, it is extremely difficult to generate the set of data necessary to train neural networks – it must include facial micromovements that even experts cannot detect. This was told socialbites.ca by Mikhail Baev, a psychotherapist who specializes in the analysis of facial expressions.

“The appearance of expressions of true emotions on a person’s face is a rare occurrence, no more than 1-2 times an hour. The rest of the time we see a fragmented facial expression. Facial expressions are much more complex and not just emotions. After analysis, neural networks give the result as a percentage, for example: 10% anger, 20% contempt, and 5% joy. But what conclusion can be drawn from such an analysis? It contradicts the concept of expressing emotions,” he explained.

Also, facial movements can be very fast – 200 to 500 milliseconds. To train neural networks, you need a corpus of photos or videos where experts mark these micromovements but a person cannot detect them. Accordingly, creating the necessary data array will not work.

For more information on how this problem was solved by the creators of the “digital profiler” from Moscow State University and in what areas it can be used – material “socialbites.ca”.

No time to read?
Get a summary
Previous Article

Steve Jobs’ rare signature is on sale for $95,000

Next Article

Scientists find a link between daytime sleepiness and dementia