Qianli Ma

Chaotic Time Series Prediction Based on Evolving Recurrent Neural Networks

The prediction of future values of a time series generated by a chaotic dynamical system is a challenging task. Recently, the use of recurrent neural networks (RNN) models appears. An evolving neural network (ERNN) is proposed for the prediction of chaotic time series, which estimates the proper parameters of phase space reconstruction and optimizes the structure of recurrent neural networks by evolutionary algorithms. The effectiveness of ERNN is evaluated by using four benchmark chaotic time series data sets: Lorenz series, logistic series, Mackey-Glass series and real-world sun spots series. Our experiments indicate that the prediction performances of ERNN are better than the other methods exiting in the bibliography.

Enhanced locality preserving projections using robust path based similarity

Curse of dimensionality is a bothering problem in high dimensional data analysis. To enhance the performances of classification or clustering on these data, their dimensionalities should be reduced beforehand. Locality Preserving Projections (LPP) is a widely used linear dimensionality reduction method. It seeks a subspace in which the neighborhood graph structure of samples is preserved. However, like most dimensionality reduction methods based on graph embedding, LPP is sensitive to noise and outliers, and its effectiveness depends on choosing suitable parameters for constructing the neighborhood graph. Unfortunately, it is difficult to choose effective parameters for LPP. To address these problems, we propose an Enhanced LPP (ELPP) using a similarity metric based on robust path and a Semi-supervised ELPP (SELPP) with pairwise constraints. In comparison with original LPP, our methods are not only robust to noise and outliers, but also less sensitive to parameters selection. Besides, SELPP makes use of pairwise constraints more efficiently than other comparing methods. Experimental results on real world face databases confirm their effectiveness.

Adaptive semi-supervised dimensionality reduction based on pairwise constraints weighting and graph optimizing

With the rapid growth of high dimensional data, dimensionality reduction is playing a more and more important role in practical data processing and analysing tasks. This paper studies semi-supervised dimensionality reduction using pairwise constraints. In this setting, domain knowledge is given in the form of pairwise constraints, which specifies whether a pair of instances belong to the same class (must-link constraint) or different classes (cannot-link constraint). In this paper, a novel semi-supervised dimensionality reduction method called adaptive semi-supervised dimensionality reduction (ASSDR) is proposed, which can get the optimized low dimensional representation of the original data by adaptively adjusting the weights of the pairwise constraints and simultaneously optimizing the graph construction. Experiments on UCI classification and image recognition show that ASSDR is superior to many existing dimensionality reduction methods.

Functional echo state network for time series classification

Abstract Echo state networks (ESNs) are a new approach to recurrent neural networks (RNNs) that have been successfully applied in many domains. Nevertheless, an ESN is a predictive model rather than a classifier, and methods to employ ESNs in time series classification (TSC) tasks have not yet been fully explored. In this paper, we propose a novel ESN approach named functional echo state network (FESN) for time series classification. The basic idea behind FESN is to replace the numeric variable output weights of an ESN with time-varying output-weight functions and introduce a temporal aggregation operator to the output layer that can project temporal signals into discrete class labels, thereby transforming the ESN from a predictive model into a true classifier. Subsequently, to learn the output-weight functions, a spatio-temporal aggregation learning algorithm is proposed based on orthogonal function basis expansion. By leveraging the nonlinear mapping capacity of a reservoir and the accumulation of temporal information in the time domain, FESN can not only enhance the separability of different classes in a high-dimensional functional space but can also consider the relative importance of temporal data at different time steps according to dynamic output-weight functions. Theoretical analyses and experiments on an extensive set of UCR data were conducted on FESN. The results show that FESN yields better performance than single-algorithm methods, has comparable accuracy with ensemble-based methods and exhibits acceptable computational complexity. Interestingly, for some time series datasets, we visualized some interpretable features extracted by FESN via specific patterns within the output-weight functions.

Improving diversity in Web search results re-ranking using absorbing random walks

Search result diversification has become important for improving Web search effectiveness and user satisfaction, as redundancy in top ranking results often disappoints users. To solve this problem, many techniques have been proposed to make a tradeoff between the relevance and diversity. Among them, GRASSHOPPER which utilizes the framework of absorbing random walks has shown good performance. In this paper, we propose a novel algorithm named DATAR with a new ranking strategy, which improves the diversification ability of GRASSHOPPER. Also, we make a discussion on the reason why DATAR is better. We evaluated the proposed algorithm with a public dataset ODP239 and a real search result dataset collected from Google. The experiment results show that the proposed DATAR algorithm outperforms GRASSHOPPER in improving diversity in Web search results re-ranking.

Mixture graph based semi-supervised dimensionality reduction

Graph structure is crucial to graph based dimensionality reduction. A mixture graph based semi-supervised dimensionality reduction (MGSSDR) method with pairwise constraints is proposed. MGSSDR first constructs multiple diverse graphs on different random subspaces of dataset, then it combines these graphs into a mixture graph and does dimensionality reduction on this mixture graph. MGSSDR can preserve the pairwise constraints and local structure of samples in the reduced subspace. Meanwhile, it is robust to noise and neighborhood size. Experimental results on facial images feature extraction demonstrate its effectiveness.

The research of the parallel SMO algorithm for solving SVM

In order to improve solving Support Vector Machine algorithm, an improved learning algorithm of the parallel SMO is proposed. According to this algorithm, the master CPU averagely distributes primitive training set to slave CPUs so that they can almost independently run serial SMO on their respective training set. As it adopts the strategies of buffer and shrink, the speed of the parallel training algorithm is increased, which is showed in the experiments of parallel SMO based on the dataset of MNIST. The experiments indicate that the parallel SMO algorithm has good performance in solving largescale SVM.

Robust path based semi-supervised dimensionality reduction

In many pattern recognition and data mining tasks, we often confront the problem of learning from a large amount of unlabeled data only with few pairwise constraints. This learning style is a kind of semi-supervised learning, and these pairwise constraints are called Side-Information. Generally speaking, these pairwise constraints are divided into two categories, one is called must-link if the pair of instances belongs to the same class, and the other is called cannot-link if the pair of instances belongs to different classes. Curse of dimensionality comes out simultaneously when the original data space is high, thus, many dimensionality reduction algorithms have proposed, and some of them utilize the side-information of the samples. However, the best learning result cannot be achieved only by using the side-information. So, we propose a novel algorithm called Robust Path Based Semi-Supervised Dimensionality Reduction (RPSSDR) in this paper. The proposed RPSSDR can not only utilize the pairwise constraints but also capture the manifold structure of the data by using robust path based similarity measure. A kernel extension of RPSSDR for the nonlinear dimensionality reduction is also presented. Besides, it can get a transformation matrix and handle unseen sample easily. Experimental results on high dimensional facial databases prove the effectiveness of our proposed method.

A Keyword Based Prototype for Web Search Result Diversification

In web search scenario, users often submit short query terms to search engines, expecting to find their desired information in top ranked results. But their queries are so ambiguous that their actual information needs are often unspecified. To satisfy the different information needs, an effective approach is to diversify the top results retrieved for the query. In this paper, we reduce the diversification problem into optimizing the maximum coverage of information facets related to the query, and introduce KED, a novel keyword based prototype for web search result diversification that provides a diverse ranking by selecting documents to cover keywords which belong to different facets underlying the retrieved documents. We evaluated the effectiveness of KED using two public test collections with different kinds of documents. The experiment results show that KED can stably outperform other existing implicit diversification approaches in promoting diversity of top ranked results. Moreover, we show that its effectiveness can be further improved by using high quality keywords.

Shared reservoir modular echo state networks for chaotic time series prediction

This paper proposes a new RNN — shared reservoir modular echo-state networks (SRMESNs), which has a higher forecast precision when the amount of training data is large enough. First, the neural state space is divided into several subspaces. And then the data belonging to each subspace is put into the same reservoir. But for each subspace, we set up an independent output weight vector respectively. So it combines the advantages of ESNs and modularization. The method is tested on the benchmark prediction problem of Mackey-Glass time series, and the result shows that the methodology proposed is efficient.