Efficient coding in active perception: A developmental perspective on autonomous control.

Area of Science:

• Developmental Psychology
• Computational Neuroscience
• Information Theory

Background:

• The efficient coding hypothesis posits that perceptual systems adapt to environmental sensory statistics.
• Infant perception and action are linked, as infants actively shape sensory input through behavior.
• Existing models need extension to explain the development of active perception.

Purpose of the Study:

• To review extensions of efficient coding to active perception.
• To focus on the Active Efficient Coding (AEC) framework.
• To provide a unifying computational principle for perception and action development.

Main Methods:

• Reviewing recent extensions of efficient coding into active perception.
• Introducing a novel formalism framing AEC within rate-distortion theory.
• Re-examining AEC models for active binocular vision in simulated infant embodiment.

Main Results:

• Active perception is interpreted as a lossy compression problem within rate-distortion theory.
• AEC models demonstrate autonomous learning and calibration of active binocular vision.
• Active perception can emerge without external supervision.

Conclusions:

• The Active Efficient Coding (AEC) framework offers a unifying principle for perception and action development.
• Active perception development is grounded in efficient sensory data compression.
• This framework provides a foundation for understanding complex behaviors and higher cognition in infants.