Developmental changes in exploration resemble stochastic optimization.

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Neuroscience

Background:

• Human development is often analogized to a 'cooling off' process, similar to stochastic optimization algorithms reducing randomness over time.
• Existing research lacks empirical comparisons to clarify this analogy, leading to ambiguity in understanding developmental changes.
• The analogy may oversimplify development by focusing solely on randomness reduction.

Purpose of the Study:

• To empirically investigate the 'cooling off' analogy in human development beyond just randomness.
• To compare human learning parameter development with computational optimization algorithms.
• To identify similarities and differences in developmental trajectories and convergence patterns.

Main Methods:

• Analyzed data from 281 participants aged 5 to 55.
• Examined the developmental trajectories of multiple learning parameters, including reward generalization, uncertainty-directed exploration, and random temperature.
• Compared human developmental data with several stochastic optimization algorithms.

Main Results:

• Human development involves optimizing multiple learning parameters, not solely reducing randomness.
• Parameters change rapidly in childhood and stabilize to efficient values in adulthood.
• Human developmental trajectories resemble stochastic optimization algorithms but show distinct convergence patterns.

Conclusions:

• Human development is a multi-parameter optimization process, extending beyond simple 'cooling off'.
• Adults exhibit efficient learning strategies comparable to, and in some cases exceeding, tested computational models.
• The study highlights unique aspects of human developmental convergence compared to artificial optimization.