Toward a self-organizing pre-symbolic neural model representing sensorimotor primitives

Area of Science:

• Cognitive Science
• Robotics
• Computational Neuroscience

Background:

• Symbolic and linguistic representations are crucial for sensorimotor behavior, developing during early cognitive stages per Piaget's theory.
• Understanding how agents acquire these representations from sensory input is key to artificial intelligence and cognitive modeling.

Purpose of the Study:

• To propose and exemplify a computational model linking visual stimuli conceptualization to the development of ventral/dorsal visual streams.
• To demonstrate how neural networks can self-organize pre-symbolic representations from observed sensorimotor data.

Main Methods:

• A neural network architecture was developed, featuring a predictive sensory module (Recurrent Neural Network with Parametric Biases) and a horizontal product model.
• A robot passively observed an object, learning its features and movement trajectories.
• Analysis focused on the self-organization of pre-symbolic representations within parametric units during sensorimotor primitive learning.

Main Results:

• The model successfully self-organized pre-symbolic representations in parametric units during sensorimotor primitive learning.
• These representations acted as bifurcation parameters, enabling the robot to recognize and predict learned sensorimotor primitives.
• The pre-symbolic representation framework also explained latent learning of sensorimotor primitives.

Conclusions:

• The proposed model provides a computational framework for understanding the development of symbolic representations from sensorimotor experiences.
• This approach links visual processing streams to cognitive development theories and demonstrates emergent representational capabilities in AI agents.
• The findings have implications for artificial general intelligence and understanding biological cognitive development.