Brief category learning distorts perceptual space for complex scenes.

Area of Science:

• Cognitive Psychology
• Neuroscience
• Computer Vision

Background:

• Category formation is known to distort perceptual space, influencing how we represent and recall information.
• Previous studies focused on artificial objects, leaving distortions for complex, real-world scenes largely unexplored due to measurement challenges.

Purpose of the Study:

• To investigate how category learning induces perceptual distortions in complex, real-world scene perception.
• To explore the impact of arbitrary category boundaries on memory reconstruction of scenes.

Main Methods:

• Generated realistic scene images using generative adversarial networks (GANs) from a high-dimensional continuous space.
• Employed a novel learning task where participants categorized these scene images.
• Assessed memory reconstruction biases by having participants recall scenes after categorization training.

Main Results:

• Systematic biases in scene reconstruction errors were observed.
• Reconstruction errors closely correlated with participants' subjective category boundaries.
• Perception of global scene properties was significantly warped by the newly learned category structure.

Conclusions:

• Category-induced perceptual distortions occur even for complex, real-world scenes after brief learning.
• The brain adapts global scene perception to align with learned categorical structures.
• This highlights the dynamic and adaptive nature of human perception in organizing complex visual information.