A Unifying Computational Framework for Teaching and Active Learning.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Computational Neuroscience

Background:

• Traditional learning models focus on passive information acquisition or active environmental exploration.
• Recent models explore learning from teachers, emphasizing reasoning about evidence selection.
• A gap exists in unifying active learning and teaching frameworks.

Purpose of the Study:

• To introduce a computational framework unifying active learning and teaching.
• To propose a Self-Teaching model based on meta-reasoning applied to oneself.
• To formalize active learning within a broader teaching paradigm.

Main Methods:

• Developed a computational framework integrating meta-reasoning for self-directed learning.
• Modeled the Self-Teaching agent's behavior using simulation experiments.
• Analyzed model behavior across three established learning problems.

Main Results:

• The Self-Teaching model successfully integrates information-gain and hypothesis-testing active learning.
• The model demonstrates a formalization of active learning within a teaching framework.
• Simulation results characterize the model's behavior in controlled learning scenarios.

Conclusions:

• Self-Teaching offers a unified approach to active learning and teaching.
• Meta-reasoning is key to understanding oneself and others in learning.
• The framework provides insights into theory-of-mind-based learning and cognitive development.