CFP: Signal Processing Special Issue on Big Data Meets Multimedia Analytics

Special Issue on Big Data Meets Multimedia Analytics

With the rapid development of computing and sensing technologies, such as the emergence of social networking websites and wearable devices, many new research opportunities and challenges for multimedia content analysis have arisen.

Many big data modeling methods, computing algorithms, and signal processing technologies have recently been successfully developed and applied to multimedia content analysis: for example, multi-view learning algorithms have been proposed for exploring the variety of multimedia content; sparse and manifold learning have been developed for high dimensional multimedia data representation; deep learning has produced promising results in large scale multimedia retrieval; and compressive sensing and new sampling schemes have been investigated for big data analytics.

Motivated by the inclination to collect a set of recent advances and results in these related topics, provide a platform for researchers to exchange their innovative ideas on big modeling and computing solutions for multimedia content analytics, and introduce interesting utilizations of modeling and computing algorithms for particular social/personal media applications, this special issue will target emergent big modeling and computing methods for multimedia signal processing and understanding (with a special focus on social media and personal data).

To summarize, this special issue welcomes a broad range of submissions on the development and use of artificial intelligence and computing techniques for multimedia analytics. We are especially interested in: 1) theoretical advances as well as algorithm developments in big data technology for specific social/personal media analytics problems; 2) reports of practical applications and system innovations in social/personal media analytics; and 3) novel datasets as test beds for new developments, preferably with implemented standard benchmarks. The following list suggests (but is not limited to) possible topics of interest:

• Big Data Technology Specifically for Multimedia Analytics
• Big Data Technology for Multimedia Annotation, Tagging and Classification
• Big Data Technology for Multimedia Abstraction and Summarization
• Big Data Technology for Multimedia Indexing and Retrieval
• Big Data Technology and Computing for Social Media Analytics
• Big Data Technology and Computing for Biological Data
• Big Data Technology and Computing for Personal Data Mining
• Modeling of Wearable Device Sensor Streams
• Personal Data based Social Network Analysis and Web Mining
• Cloud Computing for Social Intelligence and Personal Data
• Deep Learning for Social Media Analytics
• Deep Learning for Security in Social Media

Important dates:

Manuscript Submission: May 01, 2015
Initial Decision: August 01, 2015
R1 Version: October 01, 2015
Acceptance Notification: November 01, 2015
Final Manuscripts Due: November 15, 2015
Anticipated Publication: January 01, 2016

Submission:

Manuscripts (Please follow Signal Processing publishing format, details can be found athttp://www.elsevier.com/ journals/signal-processing/0165-1684/guide-for-authors) should be submitted via the Electronic Editorial System of Elsevier: http://ees.elsevier.com/sigpro/. Please make sure to select the “SI: BDMA” as Article Type during the submission process.

Guest Editors:

Professor Tat-Seng Chua
School of Computing
National University of Singapore
Email: chuats@comp.nus.edu.sg

Professor Xiangjian He
Centre for Quantum Computation and Intelligent Systems
University of Technology, Sydney
Email: xiangjian.he@uts.edu.au

Professor Weifeng Liu
College of Information and Control Engineering
China University of Petroleum
Email: liuwf@upc.edu.cn

Professor Massimo Piccardi,
Faculty of Engineering and Information Technology
University of Technology, Sydney
Email: massimo.piccardi@uts.edu.au

Professor Yonggang Wen
School of Computer Engineering
Nanyang Technological University
Email: ygwen@ntu.edu.sg

Professor Dacheng Tao
Centre for Quantum Computation and Intelligent Systems
University of Technology, Sydney
Email: dacheng.tao@uts.edu.au

One paper has been accepted on MMM2015.

Hessian regularized sparse coding for human action recognition

Abstract. With the rapid increase of online videos, recognition and search in videos becomes a new trend in multimedia computing. Action recognition in videos thus draws intensive research concerns recently. Second, sparse represen-tation has become state-of-the-art solution in computer vision because it has sev-eral advantages for data representation including easy interpretation, quick index-ing and considerable connection with biological vision. One prominent sparse representation algorithm is Laplacian regularized sparse coding (LaplacianSC). However, LaplacianSC biases the results toward a constant and thus results in poor generalization. In this paper, we propose Hessian regularized sparse coding (HessianSC) for action recognition. In contrast to LaplacianSC, HessianSC can well preserve the local geometry and steer the sparse coding varying linearly along the manifold of data distribution. We also present a fast iterative shrink-age-thresholding algorithm (FISTA) for HessianSC. Extensive experiments on human motion database (HMDB51) demonstrate that HessianSC significantly outperforms LaplacianSC and the traditional sparse coding algorithm for action recognition.

Keywords: Action recognition; sparse coding; Hessian regularization; manifold learning

Hessian-Regularized Co-Training for Social Activity Recognition

Abstract

Co-training is a major multi-view learning paradigm that alternately trains two classifiers on two distinct views and maximizes the mutual agreement on the two-view unlabeled data. Traditional co-training algorithms usually train a learner on each view separately and then force the learners to be consistent across views. Although many co-trainings have been developed, it is quite possible that a learner will receive erroneous labels for unlabeled data when the other learner has only mediocre accuracy. This usually happens in the first rounds of co-training, when there are only a few labeled examples. As a result, co-training algorithms often have unstable performance. In this paper, Hessian-regularized co-training is proposed to overcome these limitations. Specifically, each Hessian is obtained from a particular view of examples; Hessian regularization is then integrated into the learner training process of each view by penalizing the regression function along the potential manifold. Hessian can properly exploit the local structure of the underlying data manifold. Hessian regularization significantly boosts the generalizability of a classifier, especially when there are a small number of labeled examples and a large number of unlabeled examples. To evaluate the proposed method, extensive experiments were conducted on the unstructured social activity attribute (USAA) dataset for social activity recognition. Our results demonstrate that the proposed method outperforms baseline methods, including the traditional co-training and LapCo algorithms.

• DOI: 10.1371/journal.pone.0108474