With the development of information technology, online social networks grow dramatically. They now play a significant role in people's life, especially for younger generation. While huge amount is available networks, privacy concerns arise. Among various protection proposals, notions as control and boundary have been introduced. Commercial networking sites adopted concept to implement mechanisms such Google circles Facebook custom lists. However, functions are not widely accepted by users,...

Semi-supervised multi-view feature learning (SMFL) is a feasible solution for webpage classification. However, how to fully extract the complementarity and correlation information effectively under semi-supervised setting has not been well studied. In this paper, we propose individual sharable (SMISFL) approach, which jointly learns multiple view-individual transformations one transformation explore view-specific property each view common across views. We design similarity preserving term,...

Traditional anomaly detectors examine a single view of instances and cannot discover multi-view anomalies, i.e., that exhibit inconsistent behaviors across different views. To tackle the problem, several have been developed recently, but they are all transductive unsupervised thus may suffer some challenges. In this paper, we propose novel inductive semi-supervised Bayesian detector. Specifically, first present generative model for normal data. Then, build hierarchical model, by assigning...

Automatic social circle detection in ego-networks is a fundamentally important task for network analysis. So far, most studies focused on how to detect overlapping circles or based both structure and node profiles. This paper asks an orthogonal research question: by leveraging multiple views of the structure? As first step, we crawl ego networks from Twitter model them six views, including user relationships, interactions, content. We then apply standard our modified multi-view spectral...

In music classification tasks, Convolutional Recurrent Neural Network (CRNN) has achieved state-of-the-art performance on several data sets. However, the current CRNN technique only uses RNN to extract spatial dependency of signal in its time dimension but not frequency dimension. We hypothesize latter can be additionally exploited improve performance. this paper, we propose an improved called Time and Frequency dimensions (CRNN-TF), which captures dependencies both multiple directions....

Fair machine learning has become a significant research topic with broad societal impact. However, most fair methods require direct access to personal demographic data, which is increasingly restricted use for protecting user privacy (e.g. by the EU General Data Protection Regulation). In this paper, we propose distributed framework of data. We assume data privately held third party, can communicate center (responsible model development) without revealing information. principled approach...

Fair machine learning is a topical problem. It studies how to mitigate unethical bias against minority people in model prediction. A promising solution ensemble - Nina et al [1] first argue that one can obtain fair by bagging set of standard models. However, they do not present any empirical evidence or discuss effective strategy for learning. In this paper, we propose new adopts the AdaBoost framework, but unlike upweights mispredicted instances, it unfairly predicted instances which...

The key of color face recognition technique is how to effectively utilize the complementary information between components and remove their redundancy. Present methods generally reduce correlations in image pixel level, then extract discriminant features from uncorrelated images. In this paper, we propose a novel approach based on holistic orthogonal analysis (HOA) transforms HOA can correlation feature level. It turn achieves red, green blue images by using Fisher criterion, simultaneously...

Manifold structure is important for a data set, and many subspace learning methods tend to preserve this in the process. In paper, we simultaneously consider distances angles between image vectors measure similarities, hope of more sufficiently capturing manifold structure. order highlight distinctions among different data, enhance complementary information compared with distance, propose new type angle measurement shifted space that centered at mean. Both distance are fused using parallel...

Sparse representation has been extensively studied in the signal processing community, which shows that one target sample can be accurately recovered by a sparse linear combination of overall data. Such discovery soon applied to pattern recognition task and, more recently, given rise two new feature extraction methods, namely sparsity preserving projections (SPP) and global (GSRP). However, both methods utilized simply it embedded space, but none them investigates its natural discriminative...

Sparse representation technique has been successfully employed to solve face recognition task. Though current sparse based classifier proves achieve high classification accuracy, it implicitly assumes that the losses of all misclassifications are same. However, in many real-world applications, different could lead losses. Driven by this concern, we propose paper a cost-sensitive for recognition. Our approach uses probabilistic model estimate posterior probabilities testing sample, calculates...

In semi-supervised multi-view learning, unlabeled sample complexity (u.s.c.) specifies the size of training that guarantees a desired learning error. this paper, we improve state-of-art u.s.c. from O(1/ε) to O(log 1/ε) for small error ε, under mild conditions. To obtain improved result, as primary step prove connection between generalization classifier and its incompatibility, which measures fitness distribution. We then with sufficiently large sample, one is able find classifiers low...

We observe standard transfer learning can improve prediction accuracies of target tasks at the cost lowering their fairness -- a phenomenon we named discriminatory transfer. examine hypothesis algorithm and multi-task algorithm, show they both suffer on real-world Communities Crime data set. The presented case study introduces an interaction between learning, as extension existing studies that focus single task learning.

To evaluate prediction qualities of machine learning models, it is typically assumed testing samples are labeled. However, labels not always available in practice. A traditional solution to approximate on by the labeled training samples. But this may be limited that completely ignores In paper, we present a new approach estimate unlabeled sample, based reverse framework [1]. We with various quality metrics classification and anomaly detection tasks, over numerous real-world data sets....

When extracting discriminative features from multimodal data, current methods rarely concern themselves with the data distribution. In this paper, we present an assumption that is consistent viewpoint of discrimination, is, a person’s overall biometric should be regarded as one class in input space, and his different can form Gaussians distributions, i.e., subclasses. Hence, propose novel feature extraction recognition approach based on subclass discriminant analysis (SDA). Specifically,...

Manifold learning is an effective feature extraction technique, which seeks a low-dimensional space where the manifold structure, in terms of local neighborhood, data set can be well preserved. A typical method constructs neighborhood centered at individual samples. In this paper, we propose to construct neighborhoods that subclass centers, and seek embedded such We show from probability perspective that, neighbors center would contain more intra-class than inter-class data, may desirable...