Meta-cognitive learning algorithm for neuro-fuzzy inference systems

Neuro-fuzzy systems are learning machines that employ algorithms derived from artificial neural networks to find the parameters of a fuzzy inference system. These hybrid-intelligent systems can learn fuzzy rules from the data, while preserving their semantic properties. The learning algorithms employed in these systems are inspired from the physiology and functioning of human brain. As the study on neuroscience and cognitive psychology is an ongoing process, it is essential that these machine learning algorithms be extended with developments from the findings in human learning psychology. Studies in human learning psychology have shown that self-regulated learning in a meta-cognitive framework is the best learning strategy. Meta-cognition also provides a learner with self-monitoring, which is a step-by-step process of evaluation during the learning process. In this thesis, we develop a Meta-Cognitive sequential learning algorithm for Neuro-Fuzzy Inference System called McFIS which is based on a well-known model of meta-cognition proposed by Nelson and Narens. Similar to Nelson and Narens model of meta-cognition, McFIS consists of a cognitive and a meta-cognitive component. A Takagi-Sugeno-Kang type neuro-fuzzy inference system forms the cognitive component and a self-regulatory learning mechanism is its meta-cognitive component. The meta-cognitive component monitors the knowledge in the cognitive component and controls the learning by efficiently deciding on which sample to learn, when to learn it and how to learn it, efficiently. Such a learning strategy helps the network generalize the functional relationship between input and output. The performance evaluation of McFIS on a set of benchmark function approximation and time-series prediction problems indicate significant improvement over other state-of-the-art techniques. In addition, quantitative study on a set classification problems indicate motivating results. It has been shown in literature that algorithms developed to solve function approximation problems may not solve classification problems efficiently. For solving classification problems, class-specific information plays a vital role. Hence, McFIS has been extended to solve classification problems, efficiently, by considering class-specific information. Here, the meta-cognitive component monitors the class-specific knowledge in the cognitive component and decides on efficient control strategy. Moreover, the learning algorithm also considers class-specific information while learning the fuzzy rules. This helps the network learn the classification decision boundary efficiently. The performance of McFIS based classifier is studied on a set of binary and multi-category classification problems. The performance comparison indicates superior classification ability of McFIS based classifier over other classifiers considered in literature. The proposed McFIS based classifier is used to solve two important problems in video analytics: action recognition and emotion recognition. The problem of human action recognition and emotion recognition are immensely difficult due to lack of features to accurately represent various actions/emotions, and lack of efficient classifiers. In the literature of action recognition, it has been shown that optical flow based features can capture apparent movement in object naturally. Moreover, they can represent information hierarchically from local pixel level to global object level. Hence, we develop McFIS based action recognition system, where the aim of McFIS is to find the optimal decision boundary separating different actions, based on their optical flow based features. We then propose a 3D action recognition system. Here, 3D optical flow features, which are combination of 2D optical flow based features with depth flow features, are employed. The use of 3D video sequence help in avoiding issues such as occlusion and view point variation. The performance of McFIS based action recognition system is evaluated on a set of benchmark as well as proprietary action recognition data sets. The performance analysis against other features and techniques indicate superior performance of McFIS with optical flow features. Further, we extend McFIS for human emotion recognition problems. From two benchmark datasets, we have extracted pixel based, Gabor filter, local binary pattern, curvelet and scale invariant feature transform based features, in order to study their effect on emotion recognition. Based on the extracted data, we perform a stratified 10-fold cross-validation study. We also conduct person independent and data set independent studies to validate effectiveness of the proposed emotion recognition system. The statistical performance study clearly indicates the emotion recognition ability of McFIS based classifier over other results reported in the literature. In the future, meta-cognitive learning mechanism will be extended to Type-2 neuro-fuzzy inference systems. Moreover, use of other models of meta-cognition will also be explored.DOCTOR OF PHILOSOPHY (SCE