Meta-learning Scheme Research Articles

Meta-learning within Projective Simulation

Learning models of artificial intelligence can nowadays perform very well on a large variety of tasks. However, in practice different task environments are best handled by different learning models, rather than a single, universal, approach. Most non-trivial models thus require the adjustment of several to many learning parameters, which is often done on a case-by-case basis by an external party. Meta-learning refers to the ability of an agent to autonomously and dynamically adjust its own learning parameters, or meta-parameters. In this work we show how projective simulation, a recently developed model of artificial intelligence, can naturally be extended to account for meta-learning in reinforcement learning settings. The projective simulation approach is based on a random walk process over a network of clips. The suggested meta-learning scheme builds upon the same design and employs clip networks to monitor the agent's performance and to adjust its meta-parameters "on the fly". We distinguish between "reflexive adaptation" and "adaptation through learning", and show the utility of both approaches. In addition, a trade-off between flexibility and learning-time is addressed. The extended model is examined on three different kinds of reinforcement learning tasks, in which the agent has different optimal values of the meta-parameters, and is shown to perform well, reaching near-optimal to optimal success rates in all of them, without ever needing to manually adjust any meta-parameter.

Vehicle identification by improved stacking via kernel principal component regression

Purpose – Many applications in intelligent transportation demand accurate categorization of vehicles. The purpose of this paper is to propose a working image-based vehicle classification system. The first component vehicle detection is implemented by applying Dalal and Triggs's histograms of oriented gradients features and linear support vector machine (SVM) classifier. The second component vehicle classification, which is the emphasis of this paper, is accomplished by an improved stacked generalization. As an effective ensemble learning strategy, stacked generalization has been proposed to combine multiple models using the concept of a meta-learner. However, it was found that the well-known meta-learning scheme multi-response linear regression (MLR) for stacked generalization performs poorly on the vehicle classification. Design/methodology/approach – A new meta-learner is then proposed based on kernel principal component regression (KPCR). The stacked generalization scheme consists of a heterogeneous classifier ensemble with seven base classifiers, i.e. linear discriminant classifier, fuzzy k-nearest neighbor, logistic regression, Parzen classifier, Gaussian mixture model, multiple layer perceptron and SVM. Findings – Experimental results using more than 2,500 images from four types of vehicles (bus, light truck, car and van) demonstrated the effectiveness of the proposed approach. The improved stacked generalization produced consistently better results when compared to any of the single base classifier used and four other beta learning algorithms, including MLR, majority voting, logistic regression and decision template. Originality/value – With the seven base classifiers, the KPCR-based stacking offers a performance of 96 percent accuracy and 95 percent κ coefficient, thus exhibiting promising potentials for real-world applications.

Model-Based Development of a Meta-Learning Support System: Taking a Presentation-Based Meta-Learning Scheme as an Example

Model-Based Development of a Meta-Learning Support System: Taking a Presentation-Based Meta-Learning Scheme as an Example

An intelligent support system for presentation-based meta-learning

Here, we propose a presentation-based meta-learning scheme. Firstly, we present support functions that we embed into the system. Secondly, we conduct experiments to verify the meaningfulness of our learning scheme. These suggest that the system can stimulate learners to reflect on their learning processes. Furthermore, the scheme can stimulate learners' meta-learning communication. The results show that users tightened their criteria when evaluating their own learning processes and understanding states. It is useful for learners to facilitate changes in their learning processes.

Improving a rule evaluation support method based on objective indices

It is a key for the successes of data mining projects in practical situations to evaluate the obtained so many patterns as valuable knowledge effectively. In order to provide an effective support, we have been developing a rule evaluation support method based on the learning models of objective rule evaluation indices. In this paper, we report two improvements of this method and their evaluations. One is improved the learning algorithm selection in the proposed method by introducing a constructive meta-learning scheme. The other is improved the sorting efficiency of objective rule evaluation indices by combining them.