Ethology-Based Approximate Adaptive Learning: A Near Set Approach

Introduction

The problem considered in this paper is how learning by a machine can adapt its behaviour to changing environmental conditions to achieve a better result. The solution to this problem hearkens back to the work of ethologist Niko Tinbergen (1940, 1942, 1948, 1951, 1953, 1963), starting in the 1940s. Tinbergen (1953b) suggested that the behaviour of swarms of interacting organisms and their environment make swarms be seen as individual. Of course, the insight in Tinbergen’s work augurs later by those who were interested in adaptive learning by societies of interacting machines. The work by Tinbergen and Konrad Lorenz (1981) led to the introduction of ethology, a comparative science of behaviour. The basic idea in the proposed approach to adaptive learning is to look behaviour of an organism as episodic and to record observed behaviours ethograms. An ethogram is a tabular representation of observed behaviours. An ethogram is a tabular representation of observed behaviours during an episode. Let s_i, a_i, r_i denote the i^th state, action, reward, respectively. Reward r_i results from performing action a_i, where 0 ≤ i ≤ n for some finite, positive integer n. Each episode consists of a finite state-action-reward sequence of the form . In this chapter, adaptive learning itself is observed at the individual level as well as at the society level.

The fundamental of the proposed adaptive learning approach is the notion of perception. It was pointed out by Ewa Orłowska (1982) that an approximation space provides a formal framework for perception. This is especially important in establishing a formal basis what has come to be known as approximate adaptive learning (Lockery and Peters, 2008; Peters 2007d), which is the capstone of a new approach of machine learning based on ethology (Lockery, Peters 2007), which was based on earlier work on ethology and machine learning (see Peters 2005b; Peters, Henry, Ramanna, 2005a; Peters and Henry, 2005). It should also be noted that the solution to the ethology-based machine learning problem has been further aided by the recent introduction of near sets (see Peters 2007a, 2007b, 2007c, 2006e; Peters, Skowron, Stepaniuk 2006, 2007) and its applications (Anwar, Patnaik, 2008; Henry and Peters, 2007; Lockery and Peters 2007). A near set is a collection of objects that have matching descriptions to some degree. One set X is considered near another set X’ in the case where there is at least one x in X with a description that matches the description of x’ in X’ (Peters, 2008a; Peters, 2008b; Peters and Wasilewski, 2008; Peters, 2007b, 2007e). Near sets can be looked as an extension of the original model for rough sets introduced by Zdzislaw Pawlak (1981) during the early 1980s.