Learning to represent a multi-context environment: more than detecting changes.

Area of Science:

• Cognitive Science
• Machine Learning
• Environmental Psychology

Background:

• Accurate environmental representation is crucial for animals and humans.
• Unobservable causal structures must be inferred from observable data.
• Learning is complicated by multi-contextual environments with un-cued changes.

Purpose of the Study:

• To explore the challenges and strategies for learners in multi-contextual environments.
• To investigate how learners detect and adapt to environmental context changes.
• To propose optimal structures for retaining learned causal models.

Main Methods:

• Discussing theoretical problems and rational learner strategies.
• Reviewing existing empirical findings supporting proposed strategies.
• Advocating for hierarchical models in causal learning.

Main Results:

• Environmental context changes necessitate detecting shifts and employing new causal models.
• Learners face increased difficulty when context changes lack explicit cues.
• Hierarchical models offer an effective structure for managing multi-contextual learning.

Conclusions:

• Hierarchical models are optimal for retaining past causal models, preventing relearning.
• Adapting to un-cued environmental changes requires sophisticated causal inference strategies.
• Understanding multi-contextual learning is key to artificial and biological intelligence.