Recent Developments in Algorithmic Teaching

Abstract

Abstract. The present paper surveys recent developments in algorithmic teaching. First, the traditional teaching dimension model is recalled. Starting from the observation that the teaching dimension model sometimes leads to counterintuitive results, recently developed approaches are presented. Here, main emphasis is put on the following aspects derived from human teaching/learning behavior: the order in which examples are presented should matter; teaching should become harder when the memory size of the learners decreases; teaching should become easier if the learners provide feedback; and it should be possible to teach infinite concepts and/or finite and infinite concept classes. Recent developments in the algorithmic teaching achieving (some) of these aspects are presented and compared.

Quantifying teaching behavior in robot learning from demonstration

“…The ability to analyse teaching behaviour is critical to improving LfD performance. It can be shown that an optimal teacher can theoretically provide the minimum number of samples required to teach a learner a task, called the teaching dimension (Goldman and Kearns, 1995;Balbach and Zeugmann, 2009;Khan et al, 2011;Cakmak and Thomaz, 2014;Zhu, 2015), by providing non-i.i.d. samples to the learner which exploit the task structure and learning method employed.…”

Eliciting good teaching from humans for machine learners

“…The fast-match heuristic evaluates each example in terms of the degree that it improves the goodness-of-fit between current and true models. 3 MacGregor observed that this heuristic favors examples that are similar to both categories, while placing prototypical, pure-case examples lower in the ranking. Based on this observation he proposed the close-all heuristic, which ranks examples based on their distance to the decision boundary, as in our illustrative example above.…”

“…A helpful teacher can significantly improve the learning rate of a ML algorithm, as shown in the field of Algorithmic Teaching [3,29,15]. This field studies the teaching problem, that is, producing a set of labeled examples based on a known target concept.…”

Optimality of human teachers for robot learners

“…A good teacher can produce a data set that results in a more accurate output with less examples. Producing such a data set for an arbitrary concept is a challenging problem and has been formally studied within the field of algorithmic teaching [Balbach and Zeugmann, 2009], [Mathias, 1997], [Goldman and Kearns, 1995].…”