Human Learning of Hierarchical Graphs.

Area of Science:

• Cognitive Science
• Network Science
• Machine Learning

Background:

• Humans process sequential environmental events with statistical regularities.
• Real-world networks often exhibit hierarchical structures, posing challenges for human learning.
• Understanding human learning of hierarchical graph topologies is an open research question.

Approach:

• Investigated human learning of hierarchical graphs using computer simulations and behavioral experiments.
• Utilized the surprisal effect (slower reaction to unexpected transitions) to probe mental estimates of transition probabilities.
• Employed mean-field predictions and numerical simulations to analyze surprisal effects in hierarchical structures.

Key Points:

• Surprisal effects are more pronounced for finer-level hierarchical transitions than coarser-level ones.
• Coarser-level surprisal effects are difficult to detect with limited learning time or small sample sizes.
• Human participants (n=100) exhibited a surprisal effect at the finer hierarchy level but not the coarser level.

Conclusions:

• Human learning of hierarchical graphs is influenced by the level of detail.
• A trade-off exists in learning: better fine-level learning can impair coarser-level learning, and vice versa.
• This research provides insights into learning sequential events in hierarchical contexts and guides future neural and behavioral studies.