The local neighborhood selection plays a crucial role for most representation based manifold learning algorithms. This paper reveals that an improper selection of neighborhood for learning representation will introduce negative components in the learnt representations. Importantly, the representations with negative components will affect the intrinsic manifold structure preservation. In this paper, a local non-negative pursuit (LNP) method is proposed for neighborhood selection and non-negative representations are learnt. Moreover, it is proved that the learnt representations are sparse and convex. Theoretical analysis and experimental results show that the proposed method achieves or outperforms the state-of-the-art results on various manifold learning problems.

With the rapid growth of event-based social networks, the demand of event recommendation becomes increasingly important. Different from classic recommendation problems, event recommendation generally faces the problems of heterogenous online and offline social relationships among users and implicit feedback data. In this paper, we present a baysian probability model that can fully unleash the power of heterogenous social relations and efficiently tackle with implicit feedback characteristic for event recommendation. Experimental results on several real-world datasets demonstrate the utility of our method.

Utilizing multiple queues in Greedy Best-First Search (GBFS) has been proven to be a very effective approach to satisficing planning. Successful techniques include extra queues based on Helpful Actions (or Preferred Operators), as well as using Multiple Heuristics. One weakness of all standard GBFS algorithms is their lack of exploration. All queues used in these methods work as priority queues sorted by heuristic values. Therefore, misleading heuristics, especially early in the search process, can cause the search to become ineffective. Type systems, as introduced for heuristic search by Lelis et al, are a development of ideas for exploration related to the classic stratified sampling approach. The current work introduces a search algorithm that utilizes type systems in a new way – for exploration within a GBFS multiqueue framework in satisficing planning. A careful case study shows the benefits of such exploration for overcoming deficiencies of the heuristic. The proposed new baseline algorithm Type-GBFS solves almost 200 more problems than baseline GBFS over all International Planning Competition problems. Type-LAMA, a new planner which integrates Type-GBFS into LAMA-2011, solves 36.8 more problems than LAMA-2011.

Scripts have been proposed to model the stereotypical event sequences found in narratives. They can be applied to make a variety of inferences including fillinggaps in the narratives and resolving ambiguous references. This paper proposes the first formal frameworkfor scripts based on Hidden Markov Models (HMMs). Our framework supports robust inference and learning algorithms, which are lacking in previous clustering models. We develop an algorithm for structure andparameter learning based on Expectation Maximizationand evaluate it on a number of natural datasets. The results show that our algorithm is superior to several informed baselines for predicting missing events in partialobservation sequences.

In this paper, we study a prototype learning problem, called Median Point-Set, whose objective is to construct a prototype for a set of given point-sets so as to minimize the total Earth Mover's Distances (EMD) between the prototype and the point-sets, where EMD between two point-sets is measured under affine transformation. For this problem, we present the first purely geometric approach. Comparing to existing graph-based approaches (e.g., median graph, shock graph), our approach has several unique advantages: (1) No encoding and decoding procedures are needed to map between objects and graphs, and therefore avoid errors caused by information losing during the mappings; (2) Staying only in the geometric domain makes our approach computationally more efficient and robust to noise. We evaluate the performance of our technique for prototype reconstruction on a random dataset and a benchmark dataset, handwriting Chinese characters. Experiments suggest that our technique considerably outperforms the existing graph-based methods.

Word embedding aims to learn a continuous representation for each word. It attracts increasing attention due to its effectiveness in various tasks such as named entity recognition and language modeling. Most existing word embedding results are generally trained on one individual data source such as news pages or Wikipedia articles. However, when we apply them to other tasks such as web search, the performance suffers. To obtain a robust word embedding for different applications, multiple data sources could be leveraged. In this paper, we proposed a two-side multimodal neural network to learn a robust word embedding from multiple data sources including free text, user search queries and search click-through data. This framework takes the word embeddings learned from different data sources as pre-train, and then uses a two-side neural network to unify these embeddings. The pre-trained embeddings are obtained by adapting the recently proposed CBOW algorithm. Since the proposed neural network does not need to re-train word embeddings for a new task, it is highly scalable in real world problem solving. Besides, the network allows weighting different sources differently when applied to different application tasks. Experiments on two real-world applications including web search ranking and word similarity measuring show that our neural network with multiple sources outperforms state-of-the-art word embedding algorithm with each individual source. It also outperforms other competitive baselines using multiple sources.

The fundamental reasoning problem in RCC-8 is deciding In contrast to the synthetic RCC-8 networks that have the consistency of a set of constraints Θ, i.e., whether there been used in the literature for evaluating the aforementioned is a spatial configuration where the relations between the reasoners, the real-world networks of Table 1 are very sparse regions can be described by Θ. Traditionally in qualitative and one to two orders of magnitude larger. The labels on spatial reasoning (QSR) consistency of such sets is decided their edges contain 1 or 2 base RCC-8 relations forming a by a backtracking algorithm which optionally uses a pathconsistency disjunction. This kind of networks have not been employed algorithm as a preprocessing step for forward in any experimental evaluation of RCC-8 reasoners with the checking. In general, this problem is NPcomplete (Renz exception of (Sioutis and Koubarakis 2012) in which the network and Nebel 1999). However it has been shown in (Renz 1999) adm1 has been used. Typically, the literature focuses that there are tractable subsets of RCC-8 for which the consistency on quite smaller networks (20 to 1000 nodes) with an average problem can be decided by path-consistency. of 4 base RCC-8 relations per edge, and an average Table 1 depicts the characteristics of some real-world node degree ranging from 4 to 20. Deciding the consistency RCC-8 networks recording the topological relations between of real-world networks is a very important task. Inconsistencies administrative regions in Europe (networks nuts, might arise because their RCC-8 relations are computed adm1, and adm2) and the world (networks gadm1 and based on the geometries of geographical objects which gadm2), and the performance of the following reasoners often have not been captured correctly (e.g., overlapping geometries regarding consistency checking: Renz-Nebel01 (Renz and between two regions that in principle are externally Nebel 2001), GQR-1500 (Gantner, Westphal, and Woelfl connected). This is the case for the networks gadm1 and 2008; Westphal and Hué 2012), PPyRCC8 (Sioutis and gadm2.

Keyphrases for a document concisely describe the document using a small set of phrases. Keyphrases were previously shown to improve several document processing and retrieval tasks. In this work, we study keyphrase extraction from research papers by leveraging citation networks. We propose CiteTextRank for keyphrase extraction from research articles, a graph-based algorithm that incorporates evidence from both a document's content as well as the contexts in which the document is referenced within a citation network. Our model obtains significant improvements over the state-of-the-art models for this task. Specifically, on several datasets of research papers, CiteTextRank improves precision at rank 1 by as much as 9-20% over state-of-the-art baselines.

Face recognition has been widely studied due to its importance in various applications. However, the case that both training images and testing images are corrupted is not well addressed. Motivated by the success of low-rank matrix recovery, we propose a novel semi-supervised low-rank matrix recovery algorithm for robust face recognition. The proposed method can learn robust discriminative representations for both training images and testing images simultaneously by exploiting the classwise block-diagonal structure. Specifically, low-rank matrix approximation can handle the possible contamination of data. Moreover, the classwise block-diagonal structure is exploited to promote discrimination of representations for robust recognition. The above issues are formulated into a unified objective function and we design an efficient optimization procedure based on augmented Lagrange multiplier method to solve it. Extensive experiments on three public databases are performed to validate the effectiveness of our approach. The strong identification capability of representations with block-diagonal structure is verified.

Given a set of axis-parallel n-dimensional boxes, the q-intersection is defined as the smallest box encompassing all the points that belong to at least q boxes. Computing the q-intersection is a combinatorial problem that allows us to handle robust parameter estimation with a numerical constraint programming approach. The q-intersection can be viewed as a filtering operator for soft constraints that model measurements subject to outliers. This paper highlights the equivalence of this operator with the search of q-cliques in a graph whose boxicity is bounded by the number of variables in the constraint network. We present a computational study of the q-intersection. We also propose a fast heuristic and a sophisticated exact q-intersection algorithm. First experiments show that our exact algorithm outperforms the existing one while our heuristic performs an efficient filtering on hard problems.