A simultaneous understanding of questions and images is crucial in Visual Question Answering (VQA). While the existing models have achieved satisfactory performance by associating with key objects images, answers also contain rich information that can be used to describe visual contents images. In this paper, we propose a re-attention framework utilize for VQA task. The first learns initial attention weights calculating similarity each word-object pair feature space. Then, map reconstructed...

More and more people express their opinions on social media such as Facebook Twitter. Predictive analysis time-series allows the stake-holders to leverage this immediate, accessible vast reachable communication channel react proact against public opinion. In particular, understanding predicting sentiment change of will allow business government agencies negative design strategies dispelling rumors post balanced messages revert paper, we present a strategy building statistical models from...

Emotion analysis of on-line user generated textual content is important for natural language processing and social media analytics tasks. Most previous emotion approaches focus on identifying users’ emotional states from text by classifying emotions into one the finite categories, e.g., joy, surprise, anger fear. However, there exists ambiguity characteristic analysis, since a single sentence can evoke multiple with different intensities. To address this problem, we introduce distribution...

Exploring and mining the explosive burst of "big data" has already generated a lot innovative applications, especially recent advances AI thus produced big values to human society civilization. However, due centralized patterns data governance activities, including creation, sharing, exchange, management, analytics, tracing, accounting, potential distributed on Internet are far away from being adequately explored. The announcement protection policies/laws such as GDPR makes problem even more...

The federated learning provides an effective solution to train collaborative models over a large scale of participated Industrial Internet Things (IIoT) applications with the help global server, building intelligent life. However, is vulnerable backdoor attack from strong malicious participants. inconspicuous and may result in devastating consequences. To resist on IIoT applications, we propose filter defense that can identify inputs restore data availability by <italic...

Federated learning allows a large number of resource-constrained clients to train globally-shared model together without sharing local data. These usually have only few classes (categories) data for training, where the distribution is non-iid (not independent identically distributed). In this article, we put forward concept <i>category privacy</i> first time indicate <i>which client has</i> , which an important but ignored privacy goal in federated with Although secure aggregation protocols...

Multimodal aspect-based sentiment analysis (MABSA) aims to extract aspects from text-image pairs and recognize their sentiments. Existing methods make great efforts align the whole image corresponding aspects. However, different regions of may relate in same sentence, coarsely establishing image-aspect alignment will introduce noise (i.e., visual noise). Besides, a specific aspect can also be interfered by descriptions other textual Considering aforementioned noises, this paper proposes an...

Infrastructure resources on-demand requires resource provision (e.g., CPU and memory) to be both sufficient necessary, which is the most important issue a challenge in Cloud Computing. Platform as service (PaaS) encapsulates layer of software that includes middleware, even development environment, provides them for building deploying cloud applications. In PaaS, infrastructure management becomes more challenging due thousands applications share compete simultaneously. The fundamental...

Sentiment analysis of such opinionated online texts as reviews and comments has received increasingly close attention, yet most the work is intended to deal with detection authors’ emotion. In contrast, this article presents our study social emotion problem, objective which identify evoked emotions readers by documents news articles. A novel Latent Discriminative Model (LDM) proposed for task. LDM works introducing intermediate hidden variables model latent structure input text corpora. To...

Multimodal cross-domain sentiment analysis aims at transferring domain-invariant information across datasets to address the insufficiency of labeled data. Existing adaptation methods achieve well performance by remitting discrepancies in characteristics multiple modalities. However, expressive styles different also contain domain-specific information, which hinders performance. In this article, we propose a disentangled representation adversarial network (DiSRAN) reduce domain shift for...

Approximate Nearest neighbor search (ANNS) is fundamental and essential operation in applications from many domains, such as databases, machine learning, multimedia, computer vision. Although algorithms have been continuously proposed the literature above domains each year, there no comprehensive evaluation analysis of their performances. In this paper, we conduct a experimental state-of-the-art methods for approximate nearest search. Our study (1) cross-disciplinary (i.e., including 16...

In many online news services, users often write comments towards in subjective emotions such as sadness, happiness or anger. Knowing can help understand the preferences and perspectives of individual users, therefore may facilitate publishers to provide more relevant services users. Although building emotion classifiers is a practical task, it highly depends on sufficient training data that not easy be collected directly manually labeling work quite labor intensive. Also, has different...

Visual Question Answering~(VQA) requires a simultaneous understanding of images and questions. Existing methods achieve well performance by focusing on both key objects in words However, the answer also contains rich information which can help to better describe image generate more accurate attention maps. In this paper, utilize answer, we propose re-attention framework for VQA task. We first associate question calculating similarity each object-word pairs feature space. Then, based learned...

To determine the effect of posterior condylar offset (PCO) on clinical results after total knee arthroplasty (TKA) using a high-flex posterior-stabilized (PS) fixed-bearing prosthesis.We prospectively studied and radiographic materials 89 consecutive female patients (89 knees), who had undergone primary TKAs for end-stage osteoarthritis. All operations were performed by single senior surgeon or under his supervision same operative technique. Based corrected PCO change, we divided all cases...

With the increase of web applications in information society, Web application software security become more and important. Recent investigations show that vulnerabilities have largest threat. Websense report shows first half year 2008 above 75% most popular site utilized by hackers to run malicious code. Detecting solving vulnerability is effective way enhance security. In this paper we focus on regression test detection, present a strong-association rule based algorithm make detection...

Social emotion analysis of online users has become an important task for mining public opinions, which aims at detecting the readers' emotions evoked by news articles. In this paper, we focus on building a social system (SEAS) news. The implemented text data crawler mainstream websites, modules document preprocessing, representation, and also integrates successful methods provides corresponding performance evaluation. SEAS will automatically analyze towards certain articles output predicted...