Self-Directed Learner

Traditional supervised learning algorithms choose labeled training examples in a given sequence passively. However, in many real-world situations, a learner can choose which training example to learn, and its goal is to minimize the number of mistakes that the learner currently predicts for such training examples. In this paper, we propose a simple yet effective human-oriented supervised learning paradigm, Self-Directed Learner (SDL), which explicitly exploits a human learning strategy to solve this problem. SDL chooses the example that is predicted with the most certain label to learn and updates its model gradually. We conduct the experiments on a well-known educational software with both our learning algorithm and human beings. The experiment results show that HOL is able to minimize the number of mistakes efficiently. In addition, it models the human learning process much better than other learning algorithms.