Development of Schemata during Event Parsing: Neisser's Perceptual Cycle as a Recurrent Connectionist Network

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Human event parsing involves integrating sensory input with prior knowledge.
• Understanding how the brain processes sequential information is a fundamental cognitive challenge.
• Previous models acknowledged schema similarity but lacked mechanistic detail.

Purpose of the Study:

• To simulate human event parsing using recurrent neural networks.
• To investigate the interplay of top-down and bottom-up processing in event perception.
• To identify computational factors influencing schema development and utilization.

Main Methods:

• Utilized a recurrent neural network (RNN) trained on human-observed event sequences.
• Tested transfer performance by training on partial sequences and testing on unseen data.
• Employed cluster analysis of hidden units to examine network internal dynamics.

Main Results:

• Identified two primary factors in event perception: input-schema similarity and expected event duration.
• Demonstrated how RNNs can instantiate similarity judgments using both processing streams.
• Revealed computational patterns suggesting a mechanism for schema development within the network.

Conclusions:

• Neural network simulations offer insights into the mechanisms of human event perception.
• Similarity and temporal expectations are critical computational components of event understanding.
• This approach provides a potential model for how cognitive schemas are formed and updated.