טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
Ph.D Thesis
Ph.D StudentHupert-Graff Gil
SubjectComparative Study of Learning Processes in Human and in
Neural Networks
DepartmentDepartment of Industrial Engineering and Management
Supervisors Professor Emeritus Daniel Gopher
Ms. Hava Sigelmann


Abstract

The goal of this research is to enhance the scientific knowledge on human learning processes, focusing on coping with dynamic situations.


We developed a methodology called "Knowledge Tomography" that utilizes neural networks in order to capture, and to investigate, status of human knowledge states. We used it to research two major aspects of adaptive expertise:

Who are more flexible novices or experts?

·        Experts were shown to be more flexible than novices - they cope better with transfer tasks.

·        However, personal style which is developed during acquisition may harms the expert’s flexibility.

·        In transfer the first effect declines very slow relative to the second one.

How is adaptive expertise acquired, and what is the best training procedure to achieve it?

1.    A procedure that channels the trainee straight to the expert’s strategy.

2.    A more natural procedure that lets the trainee to converge to the expert’s strategy after longer and more diverse experience.

·        We demonstrated, with the model, the advantage of the second procedure.

·        Documenting the current behavior is not enough, in order to encapsulate expert's knowledge. One should explore all his/her acquisition process.

·        The natural acquisition process includes exploration of strategies that ought to “extinct” because they are not optimal for the learning task. But, these strategies might be reused, and found useful in future transfer tasks.

·        In order to develop adaptive expertise, one has to expose trainees during training to a diverse body of relevant strategies. The trainee should be exposed but not necessarily gain a deep understanding of the strategies used.