This paper addresses several key issues in the ArnetMiner system, which aims at extracting and mining academic social networks. Specifically, system focuses on: 1) Extracting researcher profiles automatically from Web; 2) Integrating publication data into network existing digital libraries; 3) Modeling entire network; 4) Providing search services for network. So far, 448,470 have been extracted using a unified tagging approach. We integrate publications online Web databases propose...

Influence maximization is the problem of finding a small set most influential nodes in social network so that their aggregated influence maximized. In this paper, we study linear threshold model, one important models formalizing behavior propagation networks. We first show computing exact general networks model #P-hard, which closes an open left seminal work on by Kempe, Kleinberg, and Tardos, 2003. As contrast, directed cyclic graphs (DAGs) can be done time to size graphs. Based fast...

Most existing person re-identification (re-id) methods focus on learning the optimal distance metrics across camera views. Typically a person's appearance is represented using features of thousands dimensions, whilst only hundreds training samples are available due to difficulties in collecting matched images. With number much smaller than feature dimension, thus face classic small sample size (SSS) problem and have resort dimensionality reduction techniques and/or matrix regularisation,...

Embedding based models have been the state of art in collaborative filtering for over a decade. Traditionally, dot product or higher order equivalents used to combine two more embeddings, e.g., most notably matrix factorization. In recent years, it was suggested replace with learned similarity e.g. using multilayer perceptron (MLP). This approach is often referred as neural (NCF). this work, we revisit experiments NCF paper that popularized similarities MLPs. First, show proper...

Existing fully-supervised person re-identification (ReID) methods usually suffer from poor generalization capability caused by domain gaps. The key to solving this problem lies in filtering out identity-irrelevant interference and learning domain-invariant representations. In paper, we aim design a generalizable ReID framework which trains model on source domains yet is able generalize/perform well target domains. To achieve goal, propose simple effective Style Normalization Restitution...

Structural comparison of large trees is a difficult task that only partially supported by current visualization techniques, which are mainly designed for browsing. We present TreeJuxtaposer, system to support the several hundred thousand nodes. introduce idea "guaranteed visibility", where highlighted areas treated as landmarks must remain visually apparent at all times. propose new methodology detailed structural between two and provide nearly-linear algorithm computing best corresponding...

In this paper, we propose an analysis mechanism-based structured discriminative dictionary learning (ADDL) framework. The ADDL seamlessly integrates learning, representation, and classifier training into a unified model. applied mechanism can make sure that the learned dictionaries, representations, linear classifiers over different classes are independent discriminating as much possible. is obtained by minimizing reconstruction error analytical incoherence promoting term encourages...

In this work, we propose an exemplar-based face image segmentation algorithm. We take inspiration from previous works on parsing for general scenes. Our approach assumes a database of exemplar images, each which is associated with hand-labeled map. Given test image, our algorithm first selects subset images the database, then computes nonrigid warp to align it image. Finally, propagate labels in pixel-wise manner, using trained weights modulate and combine label maps different exemplars....

We propose two nuclear- and L2,1-norm regularized 2D neighborhood preserving projection (2DNPP) methods for extracting representative image features. 2DNPP extracts features by minimizing a Frobenius norm-based reconstruction error that is very sensitive noise outliers in given data. To make the distance metric more reliable robust, encode accurately, we minimize L2,1-norm-based error, respectively measure it over each image. Technically, enhanced variants of 2DNPP, nuclear-norm-based sparse...

Recovering low-rank and sparse subspaces jointly for enhanced robust representation classification is discussed. Technically, we first propose a transductive principal feature coding (LSPFC) formulation that decomposes given data into component part encodes features noise-fitting error part. To well handle the outside data, then present an inductive LSPFC (I-LSPFC). I-LSPFC incorporates embedded by projection one problem direct minimization, so can effectively map both inside underlying to...

In recent years, object detection has shown impressive results using supervised deep learning, but it remains challenging in a cross-domain environment. The variations of illumination, style, scale, and appearance different domains can seriously affect the performance models. Previous works use adversarial training to align global features across domain shift achieve image information transfer. However, such methods do not effectively match distribution local features, resulting limited...

Numerical evaluations with comparisons to baselines play a central role when judging research in recommender systems. In this paper, we show that running properly is difficult. We demonstrate issue on two extensively studied datasets. First, results for have been used numerous publications over the past five years Movielens 10M benchmark are suboptimal. With careful setup of vanilla matrix factorization baseline, not only able improve upon reported baseline but even outperform any newly...

Most existing person re-identification (re-id) methods focus on learning the optimal distance metrics across camera views. Typically a person's appearance is represented using features of thousands dimensions, whilst only hundreds training samples are available due to difficulties in collecting matched images. With number much smaller than feature dimension, thus face classic small sample size (SSS) problem and have resort dimensionality reduction techniques and/or matrix regularisation,...

Estimating the disparity and normal direction of one pixel simultaneously, instead only disparity, also known as 3D label methods, can achieve much higher subpixel accuracy in stereo matching problem. However, it is extremely difficult to assign an appropriate each from continuous space R3 while maintaining global consistency because infinite parameter space. In this paper, we propose a novel algorithm called PatchMatch-based superpixel cut labels image more accurately. order robust precise...