We propose a heuristic approximation algorithm for the 1-median problem. The 1-median problem is the problem of finding a vertex with the highest closeness centrality. Starting from a randomly selected vertex, our algorithm repeats to find a vertex with higher closeness centrality by approximately calculating closeness centrality of each vertex using simpler spanning subgraphs, which are called k-neighbor dense shortest path graphs with shortcuts. According to our experimental results using real networks with more than 10,000 vertices, our algorithm is more than 100 times faster than the exhaustive search and more than 20 times faster than the state-of-the-art approximation algorithm using annotated information to the vertices while the solutions output by our algorithm have higher approximation ratio.

In this paper, we introduce a self-constructive Normalized Gaussian Network (NGnet) for online learning tasks. In online tasks, data samples are received sequentially, and domain knowledge is often limited. Then, we need to employ learning methods to the NGnet that possess robust performance and dynamically select an accurate model size. We revise a previously proposed localized forgetting approach for the NGnet and adapt some unit manipulation mechanisms to it for dynamic model selection. The mechanisms are improved for more robustness in negative interference prone environments, and a new merge manipulation is considered to deal with model redundancies. The effectiveness of the proposed method is compared with the previous localized forgetting approach and an established learning method for the NGnet. Several experiments are conducted for a function approximation and chaotic time series forecasting task. The proposed approach possesses robust and favorable performance in different learning situations over all testbeds.

We propose a policy UCB-SC for budgeted multi-armed bandits. The policy is a variant of recently proposed KL-UCB-SC. Unlike KL-UCB-SC, which is computationally prohibitive, UCB-SC runs very fast while keeping KL-UCB-SC's asymptotical optimality when reward and cost distributions are Bernoulli with means around 0.5, which are verified both theoretically and empirically.

Temporal Self-Similarity Matrix (SSM) based action recognition is one of the important approaches of single-person oriented action analysis in computer vision. In this study, we propose a new kind of SSM and a fast computation method. The computation method does not require time-consuming pre-processing to find bounding boxes of the human body, instead it processes difference images to obtain action patterns which can be done very quickly. The proposed SSM is experimentally confirmed to have high power/capacity to achieve a better classification performance than four typical kinds of SSMs.

For the last few decades, learning with multiple kernels, represented by the ensemble kernel regressor and the multiple kernel regressor, has attracted much attention in the field of kernel-based machine learning. Although their efficacy was investigated numerically in many works, their theoretical ground is not investigated sufficiently, since we do not have a theoretical framework to evaluate them. In this paper, we introduce a unified framework for evaluating kernel regressors with multiple kernels. On the basis of the framework, we analyze the generalization errors of the ensemble kernel regressor and the multiple kernel regressor, and give a sufficient condition for the ensemble kernel regressor to outperform the multiple kernel regressor in terms of the generalization error in noise-free case. We also show that each kernel regressor can be better than the other without the sufficient condition by giving examples, which supports the importance of the sufficient condition.

Multi-label classification is an appealing and challenging supervised learning problem, where multiple labels, rather than a single label, are associated with an unseen test instance. To remove possible noises in labels and features of high-dimensionality, multi-label dimension reduction has attracted more and more attentions in recent years. The existing methods usually suffer from several problems, such as ignoring label outliers and label correlations. In addition, most of them emphasize on conducting dimension reduction in an unsupervised or supervised way, therefore, unable to utilize the label information or a large amount of unlabeled data to improve the performance. In order to cope with these problems, we propose a novel method termed Robust sEmi-supervised multi-lAbel DimEnsion Reduction, shortly READER. From the viewpoint of empirical risk minimization, READER selects most discriminative features for all the labels in a semi-supervised way. Specifically, the ℓ2,1-norm induced loss function and regularization term make READER robust to the outliers in the data points. READER finds a feature subspace so as to keep originally neighbor instances close and embeds labels into a low-dimensional latent space nonlinearly. To optimize the objective function, an efficient algorithm is developed with convergence property. Extensive empirical studies on real-world datasets demonstrate the superior performance of the proposed method.

Vision based human action recognition has been an active research field in recent years. Exemplar matching is an important and popular methodology in this field, however, most previous works perform exemplar matching on the whole input video clip for recognition. Such a strategy is computationally expensive and limits its practical usage. In this paper, we present a martingale framework for selection of characteristic frames from an input video clip without requiring any prior knowledge. Action recognition is operated on these selected characteristic frames. Experiments on 10 studied actions from WEIZMANN dataset demonstrate a significant improvement in computational efficiency (54% reduction) while achieving the same recognition precision.

We study the budgeted multi-armed bandit problem with stochastic action costs. In this problem, a player not only receives a reward but also pays a cost for an action of his/her choice. The goal of the player is to maximize the cumulative reward he/she receives before the total cost exceeds the budget. In the classical multi-armed bandit problem, a policy called KL-UCB is known to perform well. We propose KL-UCB-SC, an extension of this policy for the budgeted bandit problem. We prove that KL-UCB-SC is asymptotically optimal for the case of Bernoulli costs and rewards. To the best of our knowledge, this is the first result that shows asymptotic optimality in the study of the budgeted bandit problem. In fact, our regret upper bound is at least four times better than that of BTS, the best known upper bound for the budgeted bandit problem. Moreover, an empirical simulation we conducted shows that the performance of a tuned variant of KL-UCB-SC is comparable to that of state-of-the-art policies such as PD-BwK and BTS.

Many image-processing techniques are based on texture features or gradation features of the image. However, Landsat images are complex; they also include physical features of reflection radiation and heat radiation from land cover. In this paper, we describe a method of constructing a super-resolution image of Band 6 of the Landsat TM sensor, oriented to analysis of an agricultural area, by combining information (texture features, gradation features, physical features) from other bands. In this method, a knowledge-based hierarchical classifier is first used to identify land cover in each pixel and then the least-squares approach is applied to estimate the mean temperature of each type of land cover. By reassigning the mean temperature to each pixel, a finer spatial resolution is obtained in Band 6. Computational results show the efficiency of this method.

Maximum inner product search (MIPS) problem has gained much attention in a wide range of applications. In order to overcome the curse of dimensionality in high-dimensional spaces, most of existing methods first transform the MIPS problem into another approximate nearest neighbor search (ANNS) problem and then solve it by Locality Sensitive Hashing (LSH). However, due to the error incurred by the transmission and incomprehensive search strategies, these methods suffer from low precision and have loose probability guarantees. In this paper, we propose a novel search method named Adaptive-LSH (AdaLSH) to solve MIPS problem more efficiently and more precisely. AdaLSH examines objects in the descending order of both norms and (the probably correctly estimated) cosine angles with a query object in support of LSH with extendable windows. Such extendable windows bring not only efficiency in searching but also the probability guarantee of finding exact or approximate MIP objects. AdaLSH gives a better probability guarantee of success than those in conventional algorithms, bringing less running times on various datasets compared with them. In addition, AdaLSH can even support exact MIPS with probability guarantee.