Tohgoroh Matsui

An Induction Algorithm Based on Fuzzy Logic Programming

This paper gives a formulation of inductive learning based on fuzzy logic programming (FLP) and a top-down algorithm for it by extending an inductive logic programming (ILP) algorithm FOIL. The algorithm was implemented and evaluated by experiments. Linguistic hedges, which modifies truth, are shown to have effect to adjust classification properties. The algorithm deals with structural domain as other ILP algorithms do and also works well with numeric attributes.

On-line Profit Sharing Works Efficiently

Reinforcement learning constructs knowledge containing state-to-action decision rules from agent’s experiences. Most of reinforcement learning methods are action-value estimation methods which estimate the true values of state-action pairs and derive the optimal policy from the value estimates. However, these methods have a serious drawback that they stray when the values for the “opposite” actions, such as moving left and moving right, are equal. This paper describes the basic mechanism of on-line profit-sharing (OnPS) which is an action-preference learning method. The main contribution of this paper is to show the equivalence of off-line and on-line in profit sharing. We also show a typical benchmark example for comparison between OnPS and Q-learning.

Adapting behavior by inductive prediction in soccer agents

This paper proposes new architecture of agent which adapts its behavior by predicting the results of actions and avoiding taking ones predicted to be failures shown in Fig. 1. Our agent consists of five parts: Observer, Planner, Actor which are same as those of basic agent and write perceiving information in the first-order formalism, Learner which acquires prediction rules using inductive logic programming (ILP), and Checker which predicts the result of an action selected by Planner and changes action if it seems to be failure. The agent classifies past actions into successes and failures, then it learns rules from them to assort the current action without information after taking it.

Detecting and revising misclassifications using ILP

This paper proposes a method for detecting misclassifications of a classification rule and then revising them. Given a rule and a set of examples, the method divides misclassifications by the rule into miscovered examples and uncovered examples, and then, separately, learns to detect them using Inductive Logic Programming (ILP). The method then combines the acquired rules with the initial rule and revises the labels of misclassified examples. The paper shows the effectiveness of the proposed method by theoretical analysis. In addition, it presents experimental results, using the Brill tagger for Part-Of-Speech (POS) tagging.

Reinforcement Learning Methods to Handle Actions with Differing Costs in MDPs

Reinforcement learning methods for environment including actions with differing costs are investigated. Through experiments we examined treatment of this problem with Q-learning, R-learning and Profit sharing. Profit sharing with a credit assignment functions considering costs is shown to have good performance in a practical sense.