• Poster Presentation:

    1. Deep Specialized Network for Illuminant Estimation
      Speaker: SHI Wu

      Abstract: Illuminant estimation to achieve color constancy is an ill-posed problem. Searching the large hypothesis space for an accurate illuminant estimation is hard due to the ambiguities of unknown reflections and local patch appearances. In this work, we propose a novel Deep Specialized Network (DS-Net) that is adaptive to diverse local regions for estimating robust local illuminants. This is achieved through a new convolutional network architecture with two interacting sub-networks, i.e. an hypotheses network (HypNet) and a selection network (SelNet). In particular, HypNet generates multiple illuminant hypotheses that inherently capture different modes of illuminants with its unique two-branch structure. SelNet then adaptively picks for confident estimations from these plausible hypotheses. Extensive experiments on the two largest color constancy benchmark datasets show that the proposed 'hypothesis selection' approach is effective to overcome erroneous estimation. Through the synergy of HypNet and SelNet, our approach outperforms state-of-the-art methods such as [1-3].

    2. 3-channel Erasure Multiple Descriptions
      Speaker: GUO Tao

      Abstract: We describe the rate-distortion region for the 3-channel Erasure Multiple Descriptions problem.

    3. Deep Markov Random Field for Image Modeling
      Speaker: WU Zhirong

      Abstract Markov Random Fields (MRFs), a formulation widely used in generative image modeling, have long been plagued by the lack of expressive power. This issue is primarily due to the fact that conventional MRFs formulations tend to use simplistic factors to capture local patterns. In this paper, we move beyond such limitations, and propose a novel MRF model that uses fully-connected neurons to express the complex interactions among pixels. Through theoretical analysis, we reveal an inherent connection between this model and recurrent neural networks, and thereon derive an approximated feed-forward network that couples multiple RNNs along opposite directions. This formulation combines the expressive power of deep neural networks and the cyclic dependency structure of MRF in a unified model, bringing the modeling capability to a new level. The feed-forward approximation also allows it to be efficiently learned from data. Experimental results on a variety of low-level vision tasks show notable improvement over state-of-the-arts.

    4. Pricing for Sharing Economy with Reputation
      Speaker: MA Qian

      Abstract: Not Provided

    5. Understanding a ride-on-demand service
      Speaker: GUO Suiming

      Abstract: Not Provided
    6. Fashion Landmark Detection in the Wild
      Speaker: LIU Ziwei

      Abstract: Visual fashion analysis has attracted much attentions in the recent years, where a crucial step towards robust fashion analysis is to identify the informative regions of clothes, such as collar and cuff. Previous work represented clothing regions by either clothes bounding boxes or human joints, which have large biases. This work presents a new task named fashion landmark detection or fashion alignment, which is to predict the positions of functional key points defined on the fashion items, such as the corners of neckline, hemline, and cuff. We study fashion alignment by proposing a new deep learning framework, which cascades multiple convolutional neural networks in three stages. These stages gradually improve the accuracy of landmark prediction with the aid of pseudo-labels and auto-routing. Benchmark dataset and baseline methods are set up for the new task. Extensive experiments demonstrate the effectiveness as well as the generalization ability of our proposed model. Fashion landmark is also compared to clothing bounding boxes and human joints in two real-world applications, fashion attribute prediction and clothes retrieval, showing that fashion landmark is a more discriminative representation in understanding fashion images.

    7. A Multidimensional Auction Mechanism for Mobile Crowdsourced Video Streaming
      Speaker: TANG Ming

      Abstract: Adaptive bitrate video streaming is a widely-used technology for mobile video streaming over HTTP. In this work, we study a crowdsourced video streaming framework, which enables nearby mobile users to crowdsource their radio resources for cooperatively adaptive bitrate video streaming. We propose a multi-dimensional auction based incentive mechanism to promote the user cooperation, supporting the asynchronous downloading and the bitrate adapting of video users. In this mechanism, each user initiates an auction whenever he is ready to download a new data segment in an asynchronous fashion, and all nearby users compete for the downloading opportunity by submitting a multidimensional bid consisting of the intended segment bitrate and the associated value. Design of such a multi-dimensional auction is very challenging, as we need to guarantee the user’s truthful reporting on the information on multiple dependent dimensions. We first propose a truthful second-score (multi-dimensional) auction framework, within which we further derive the efficient mechanism that maximizes the social welfare (of each segment downloading) and the sub-optimal mechanism that approximately maximizes the auctioneer payoff. Experiment results show that our proposed crowdsourced streaming can achieve 60%-76% of the maximum social welfare even when 80 percentage of users lose their direct network connections.

    8. Not Provided
      Speaker: YANG Chunfeng

      Abstract: In this paper, we mine and learn to predict how similar a pair of users’ interests towards videos are, based on demographic (age, gender and location) and social (friendship, interaction and group membership) information of these users. We use the video access patterns of active users as ground truth (a form of benchmark). We adopt tag-based user profiling to establish this ground truth, and justify why it is used instead of video-based methods, or many latent topic models such as LDA and Collaborative Filtering approaches.We then show the effectiveness of the different demographic and social features, and their combinations and derivatives, in predicting user interest similarity, based on different machinelearning methods for combining multiple features. We propose a hybrid tree-encoded linear model for combining the features, and show that it out-performs other linear and treebased models. Our methods can be used to predict user interest similarity when the ground-truth is not available, e.g. for new users, or inactive users whose interests may have changed from old access data, and is useful for video recommendation. Our study is based on a rich dataset from Tencent, a popular service provider of social networks, video services, and various other services in China.

    9. Image Aesthetic Assessment: An Experimental Survey
      Speaker: DENG Yubin

      Abstract: This survey aims at reviewing recent techniques used in the assessment of image aesthetic quality. The assessment of image aesthetic quality is the process of computationally distinguishing high-quality photos from low-quality ones based on photographic rules or artistic perceptions. A variety of approaches have been proposed in the literature trying to solve this challenging problem. In this survey, we present a systematic listing of the reviewed approaches based on feature types (hand-crafted features and deep features) and evaluation criteria (dataset characteristics and evaluation metrics). Main contributions and novelties of the reviewed approaches are highlighted and discussed. In addition, following the emergence of deep learning techniques, we systematically evaluate recent deep learning settings that are useful for developing a robust deep model for aesthetic scoring. Experiments are conducted using simple yet solid baselines that are competitive with the current state-of-the-arts. Moreover, we discuss the relation between image aesthetic assessment and automatic image cropping. We hope that this survey could serve as a comprehensive reference source for future research on the study of image aesthetic assessment.

    10. Toward User Mobility for OFDM-based Visible Light Communications
      Speaker: HONG Yang

      Abstract: We propose and experimentally demonstrate a mobile visible light communications (mobi-VLC) transmission system. The impact of user mobility on the performance of the mobi-VLC system is characterized and we propose the use of the channel-independent orthogonal circulant matrix transform (OCT) precoding to combat the packet loss performance degradation induced by mobility. No packet loss for 300-Mb/s transmission is achieved with 30-cm lateral moving distance at 20 cm/s.

    11. Energy-Efficient Timely Transportation of Long-Haul Heavy-DutyTrucks
      Speaker: DENG Lei

      Abstract: We consider a timely transportation problem where a heavy-duty truck travels between two locations across the national highway system, subject to a hard deadline constraint. Our objective is to minimize the total fuel consumption of the truck, by optimizing both route planning and speed planning. The problem is important for cost-effective and environment-friendly truck operation, and it is uniquely challenging due to its combinatorial nature as well as the need of considering hard deadline constraint. We first show that the problem is NP-Complete; thus exact solution is computational prohibited unless P=NP. We then design a fully polynomial time approximation scheme (FPTAS) that attains an approximation ratio of 1 + \epsilon with a network-size induced complexity of O(mn^2/\epsilon^2), where m and n are the numbers of nodes and edges, respectively. While achieving highly-preferred theoretical performance guarantee, the proposed FPTAS still suffers from long running time when applying to national-wide highway systems with tens of thousands of nodes and edges. Leveraging elegant insights from studying the dual of the original problem, we design a fast heuristic solution with O(m + n log n) complexity. The proposed heuristic allows us to tackle the energy-efficient timely transportation problem on large-scale national highway systems. We further characterize a condition under which our heuristic generates an optimal solution. We observe that the condition holds in most of the practical instances in numerical experiments, justifying the superior empirical performance of our heuristic. We carry out extensive numerical experiments using real-world truck data over the actual U.S. highway network. The results show that our proposed solutions achieve 17% (resp. 14%) fuel consumption reduction, as compared to a fastest path (resp. shortest path) algorithm adapted from common practice.

    12. Scalable Mobile Crowd Sensing via Peer-to-Peer Data Sharing
      Speaker: JIANG Changkun

      Abstract: Mobile crowdsensing (MCS) is a new paradigm of sensing by taking advantage of the rich embedded sensors of mobile user devices. However, the traditional server-client MCS architecture often suffers from the high operational cost on the centralized server (e.g., for storing and processing massive data), hence the poor scalability. Peer-to-peer (P2P) data sharing can effectively reduce the server’s cost by leveraging the user devices’ computation and storage resources. In this work, we propose a novel P2P-based MCS architecture, where the sensing data is saved and processed in user devices locally and shared among users in a P2P manner. To provide necessary incentives for users in such a system, we propose a quality-aware data sharing market, where the users who sense data can sell data to others who request data but not want to sense the data by themselves. We analyze the user behavior dynamics from the game-theoretic perspective, and characterize the existence and uniqueness of the game equilibrium. We further propose best response iterative algorithms to reach the equilibrium with provable convergence. Our simulations show that the P2P data sharing can greatly improve the social welfare, especially in the model with a high transmission cost and a low trading price.

    13. Detecting Visual Relationships with Deep Relational Networks
      Speaker: DAI Bo

      Abstract: Relationships among objects play a crucial role in image understanding. Despite the great success of deep learning techniques in recognizing individual objects, reasoning about the relationships among objects remains a challenging task. % Previous methods often treat this as a classification problem, considering each type of relationship (e.g.~``ride'') or each distinct visual phrase (e.g.~``person-ride-horse'') as a category. % Such approaches are faced with significant difficulties caused by the high diversity of visual appearance for each kind of relationships or the large number of distinct visual phrases. % We propose an integrated framework to tackle this problem. At the heart of this framework is the Deep Relational Network, a novel formulation designed specifically for exploiting the statistical dependencies between objects and their relationships. % On two large data sets, the proposed method achieves substantial improvement over state-of-the-art.

    14. Collaborative Deep Embedding via Dual Networks
      Speaker: XIONG Yilei

      Abstract: Not Provided

    15. The Rate Region of Secure Distributed Storage Systems
      Speaker: YE Fangwei

      Abstract: Not Provided

    16. Signing into One Billion Mobile App Accounts Effortlessly with OAuth2.0
      Speaker: YANG Ronghai

      Abstract: OAuth2.0 protocol has been widely adopted by mainstream Identity Providers (IdPs) to support Single-Sign-On service. Since this protocol was originally de- signed to serve the authorization need for 3rd party websites, different pitfalls have been uncovered when adapting OAuth to support mobile app authentication. To the best of our knowledge, all the attacks discovered so far, including BlackHat USA’16 [3], CCS’14 [2] and ACSAC’15 [5], require to interact with the victim, for example via malicious apps or network eavesdropping, etc. On the contrary, we have discovered a new type of widespread but incorrect usages of OAuth by 3rd party mobile app developers, which can be exploited remotely and solely by the attacker to sign into a victim’s mobile app account without any involvement/ awareness of the victim. To demonstrate the prevalence and severe impact of this vulnerability, we have developed an exploit to examine the implementations of 600 top-ranked US and Chinese Android Apps which use the OAuth2.0-based authentication service provided by three top-tier IdPs, namely Facebook, Google or Sina. Our empirical results are alarming: on average, 41.21% of these apps are vulnerable to this new attack. We have reported our findings to the affected IdPs, and received their acknowledgements/ rewards in various ways.

    17. Human Attribute Recognition by Deep Hierarchical Contexts
      Speaker: LI Yining

      Abstract: We present an approach for recognizing human attributes in unconstrained settings. We train a Convolutional Neural Network (CNN) to select the most attribute-descriptive human parts from all poselet detections, and combine them with the whole body as a pose-normalized deep representation. We further improve by using \textit{deep hierarchical contexts} ranging from human-centric level to scene level. Human-centric context captures human relations, which we compute from the nearest neighbor parts of other people on a pyramid of CNN feature maps. The matched parts are then average pooled and they act as a similarity regularization. To utilize the scene context, we re-score human-centric predictions by the global scene classification score jointly learned in our CNN, yielding final scene-aware predictions. To facilitate our study, a large-scale WIDER Attribute dataset\footnote{Dataset URL: http://mmlab.ie.cuhk.edu.hk/projects/WIDERAttribute} is introduced with human attribute and image event annotations, and our method surpasses competitive baselines on this dataset and other popular ones.

    18. Spectrum Investment with Uncertainty Based On Prospect Theory
      Speaker: YU Junlin

      Abstract:

    19. Optimal Operation of Multiple Batteries System for Primary Frequency Control
      Speaker: ZHU Diwei

      Abstract: This poster presents a partially completed work on optimal operation policy for multiple batteries system providing primary frequency control service. The model of the multiple batteries system is an extension of a work on single battery system. A recursive algorithm is designed to obtain the optimal operation policy for the batteries during the frequency regulation interval. The operation cost of providing PFC service is derived.

    20. Two-stage Spectrum Leasing Framework for Virtual Mobile Network Operators
      Speaker: ZHANG Yingxiao

      Abstract: Wireless network virtualization (WNV) allows mobile network operators (MNOs) to lease its network infrastructure or licensed spectrum to virtual mobile network operators (VMNOs). We propose a novel two-stage spectrum leasing framework to maximize the average profit of a VMNO. Specifically, a VMNO can first make long-term spectrum lease based on the prediction of the average user traffic intensity over a long time period. Besides, the VMNO can flexibly make additional short-term leases based on the actual realizations of MS locations. Then the VMNO uses the spectrum to serve its MSs according to the fading conditions. Simulation results show that the proposed two-stage spectrum leasing strategies can effectively increase the profit of VMNOs.

    21. Discover and Learn New Objects from Documentaries
      Speaker: CHEN Kai

      Abstract: Not Provided

    22. Semantic Image Labeling via Deep Parsing Network
      Speaker: LI Xiaoxiao

      Abstract: This paper addresses pixelwise image labeling by incorporating rich information into Markov Random Field (MRF), including high-order relations and mixture of label contexts. Unlike previous works that optimized MRFs using iterative algorithm, we solve MRF by proposing a Convolutional Neural Network (CNN), namely Deep Parsing Network (DPN), which enables deterministic end-to-end computation in a single forward pass. Specifically, DPN extends a contemporary CNN architecture to model unary terms and additional layers are carefully devised to approximate the mean field algorithm (MF) for pairwise terms. It has several appealing properties. First, different from the recent works that combined CNN and MRF, where many iterations of MF were required for each training image during back-propagation, DPN is able to achieve high performance by approximating one iteration of MF, because it only comprises conventional operations of CNN. Second, DPN represents various types of pairwise terms, making many existing works as its special cases. Third, DPN makes MF easier to be parallelized and speeded up in Graphical Processing Unit (GPU). DPN is thoroughly evaluated on the PASCAL VOC 2012 dataset, yielding a new state-of-the-art accuracy of 73.5% without outside training data.

    23. Rate Constrained Secret Key Agreement
      Speaker: ZHOU Qiaoqiao

      Abstract: We consider the multiterminal secret key agreement problem, where the public discussion rate is limited. Single-letter upper bounds on the maximum rate of secret key that can be generated are obtained for the general and hypergraphical source model, respectively. The latter bound is shown to be tight in the pairwise independent network (PIN) model, which gives a complete characterization of the tradeoff between the secret key rate and public discussion rate for such model.

    24. How to Bid for Tomorrow's Electricity under Demand and Market Uncertainty
      Speaker: ZHANG Ying

      Abstract: We consider the scenario where a cloud service provider (CSP) operates large-scale datacenters to provide Internet-scale service. Our objective is to minimize the electricity and bandwidth cost by optimizing electricity procurement from wholesale markets. Under the ideal setting where exact values of market prices and workloads are given, this problem is easy to solve. However, under the realistic setting where only distributions of these variables are available, the problem unfolds into a non-convex infinite-dimensional one and is challenging to solve. Our main contribution is to provide an analytical solution by exploring the full design space of strategic bidding. Trace-driven evaluations corroborate our theoretical results, demonstrate fast convergence of our algorithm, and show that it can reduce the cost for the CSP by up to 15\% as compared to baseline alternatives.

    25. vulnerability detection in smart home devices utilizing paired mobile applications
      Speaker: CHEN Jiongyi

      Abstract: With the rapid evolution of IoT (Internet of things) techniques, smart home devices (such as smart plugs and smart locks) have been widely used in our daily life. However, security problems in these devices also emerge and security analysis on embedded devices is far from being comprehensive. To detect vulnerabilities on embedded devices (including smart home devices), previous work focuses on automated analysis via emulating firmware images, but these methods either encounter difficulties in runtime compatibility or cannot even acquire the firmware images. In this proposal, I propose a novel approach to automatically identify vulnerabilities in smart home devices by sending efficient probe messages from the paired mobile applications. Since these applications provide rich information about smart home devices, we could infer protocol message formats and send effective messages for fuzz testing. My approach is expected to overcome the problem of runtime compatibility in previous work without acquisition of firmware images. Meanwhile, it is expected to detect many vulnerabilities including insufficient input check, misuse of cryptographic functions, and weak authentication

    26. The optimal randomized online algorithm for QoS buffer management
      Speaker: YANG Lin

      Abstract: The research is on the non-preemptive QoS buffer management, which has significant applications to networks providing DiffServ. In our model, each packet is assigned a profit-related value according to its QoS (Quality of Service) requirements. The assigned value can be considered as the corresponding revenue earned by the switches when the packet is buffered and delivered successfully. We assume the queues engineered in the switches are non-preemptive and work in a FIFO manner. When a new packet arrives, the switch needs to make an immediate decision on whether to buffer the incoming packet according to its value. We contribute to this problem by proposing 1) The optimal memoryless deterministic algorithm, 2) the optimal deterministic online algorithm for the fractional case, and 3) the optimal randomized online algorithm.

    27. Factor-Graph Representations of Stabilizer Quantum Codes
      Speaker: LI Xialu

      Abstract: We study normal factor graph (NFG) representations of stabilizer quantum error-correction codes (QECCs), in particular NFG representations of the stabilizer label code and the normalizer label code associated with a stabilizer QECC. The structure of the NFGs we are using is such that the (symplectic) self-orthogonality constraint that stabilizer label codes have to satisfy can be proven rather straightforwardly by applying certain NFG reformulations. We show that a variety of well-known stabilizer QECCs can be expressed in this framework: (tail-biting) convolutional stabilizer QECCs, the toric stabilizer QECCs by Kitaev, and a class of stabilizer QECCs that was recently introduced by Tillich and Z\'emor. Our approach not only gives new insights into these stabilizer QECCs, but will ultimately help to formulate new classes of stabilizer QECCs and low-complexity (approximate) decoding algorithms.

    28. Pricing-based Energy Storage Sharing and Virtual Capacity Allocation
      Speaker: ZHAO Dongwei

      Abstract: We develop a novel business model to enable virtual storage sharing among a group of users. Specifically, an aggregator owns a central physical storage unit and virtualizes the physical storage into separable virtual storage capacities that can be sold to users. Each user purchases the virtual storage capacity, and schedules the charge and discharge of the virtual storage to reduce his peak power consumption. We formulate the interaction between the aggregator and users in each operation horizon as a two-stage problem. At the beginning of the operation horizon, the aggregator first determines the unit price of virtual storage capacity to maximize her profit in Stage 1, and users decide the capacities to purchase and the storage scheduling during the operation horizon in Stage 2. Since the closed-form solution is not available and the decisions are coupled across the two stages, we characterize the solutions of the two-stage problem based on the theory of parametric linear programming. Simulation results show that compared to the case where each user acquires his own physical storage, storage virtualization reduces the overall physical capacity needed for all users by 34.9%, and the overall physical power rating by 45.1%.

    29. Scalable Estimation of Dirichlet Process Mixture Models on Distributed Data
      Speaker: WANG Ruohui

      Abstract: We consider the estimation of Dirichlet Process Mixture Models (DPMMs) in distributed environments, where data are distributed across multiple computing nodes. A key advantage of Bayesian nonparametric models such as DPMMs is that they allow new components to be introduced on the fly as needed. This, however, posts an important challenge to distributed estimation -- how to handle new components efficiently and consistently. To tackle this problem, we propose a new estimation method, which allows new components to be created locally in individual computing nodes. Components corresponding to the same cluster will be identified and merged via a probabilistic consolidation scheme. In this way, we can maintain the consistency of estimation with very low communication cost. Experiments on large real-world data sets show that the proposed method can achieve high scalability in distributed and asynchronous environments without compromising the mixing performance.

    30. Interference Alignment with Physical-Layer Network Coding in MIMO Relay Channels
      Speaker: CHAN Tse-Tin

      Abstract: This paper proposes the ideas of interference alignment with physical-layer network coding (IAPNC). While conventional interference alignment (IA) aligns interfering signals and then treats them as noise, IAPNC treats the aligned signals as combined signals and utilizes them to convey data streams to receivers so as to increase the end-to-end sum-rate. We show the ideas of IAPNC in multi-hop multiple-input multiple-output (MIMO) channels consisting of 3 transmitter-receiver pairs and 2 half-duplex decode-and-forward (DF) relays in each intermediate layer, i.e., 3-2-···-2-3 MIMO channels. We consider three users want to convey independent data streams to distinct destinations with the aid of relays. Simulation results show that IAPNC scheme in the 3-2-3 MIMO channel achieves the same degrees of freedom (DoF) as conventional IA scheme in the 3-3-3 MIMO channel which has one more relay. The results also show that the end-to-end sum-rate of IAPNC scheme outperforms that of zero-forcing (ZF) filtering scheme in medium-to-high signal-to-noise ratio (SNR) regime for the 3-2-3 MIMO channel. The performance improvement of IAPNC scheme mainly comes from efficient utilization of signals in interfering signal subspaces.

    31. Network-Coded Multiple Access with High-order Modulations
      Speaker: PAN Haoyuan

      Abstract: This poster presents the first network-coded multiple access (NCMA) system prototype operated on high-order modulations up to 16-QAM. NCMA jointly exploits physical-layer network coding (PNC) and multiuser decoding (MUD) to boost throughput of multipacket reception systems. Direct generalization of the existing NCMA decoding algorithm, originally designed for BPSK, to high-order modulations, will lead to huge performance degradation. The throughput degradation is caused by the relative phase offset between received signals from different nodes. To circumvent the phase offset problem, we investigates an NCMA system with multiple receive antennas at the access point (AP), referred to as MIMO-NCMA. We put forth a low-complexity symbol-level NCMA decoder that, together with MIMO, can substantially alleviate the performance degradation induced by relative phase offset. We implemented our designs on software-defined radio. Our experimental results show that the throughput of QPSK MIMO-NCMA is double that of both BPSK NCMA and QPSK MUD at SNR=10dB. For higher SNRs at which 16-QAM can be supported, the throughput of MIMO-NCMA can be as high as 3.5 times that of BPSK NCMA. Overall, this paper provides an implementable framework for high-order modulated NCMA.

    32. Minimizing the Maximum Number of Link Collisions in Data Center Network Routing Using Genetic Algorithms
      Speaker: TONG Kwong Bun

      Abstract: Typical Data Center Networks (DCNs) are implemented as a fat tree structure. One of the major advantages is that it has multiple equal-cost paths which provides the capability of load balancing and fault tolerant. However, to ensure a DCN achieving the highest throughput, one has to minimize the number of link collisions of the network by choosing proper routing strategies for the traffic. This work reports the first evolutionary approach for minimizing the maximum link collisions of a fat tree network. The experimental results show that it significantly reduces the chance of traffic collision in a fat tree network. It also demonstrated with the addition of a simple and easy-to-implement greedy local search, the same cost solution can be obtained earlier and the number of link collisions can be further reduced. It is observed that the return of the effort of greedy search diminishes with the drop of traffic density. It is suggested that when the traffic density is lower than certain threshold, the proposed Genetic algorithm already performs well and the greedy searches could be skipped for time critical routing.

    33. Anonymous Fingerprinting for Off-line QR Payment
      Speaker: ZHOU Zhe

      Abstract: QR code payment is gradually accepted by more and more vendors and smartphone users, as it frees people from carrying a lot of cards and charges less handling fees from vendors. However, according to our observation, each smartphone screen has its unique imaging characteristic (fingerprint) that makes the screen distinguishable by the vendor and the payment service provider during the QR code scanning process. This observation has two strong security implications: on one hand, the payer looses anonymity to vendors who original do not know who is paying; on the other, service providers who originally only check the message inside the QR code now can also authenticate the device using the fingerprint as a second factor, with which attackers who even have controlled victims' operating system and have stolen victims' payment credential can no longer pass the authentication. The story does not end here. We proposed AnonPrint that disables vendors's likability to payers while at the same time retain the second factor authentication capability at server side, by obfuscating the fingerprint at wallet application side but de-obfuscating it at server side, where the obfuscation is achieved by displaying a mask as background. With AnonPrint, payers can confidently pay with QR code without letting vendor knows who is paying. Besides, payment server does not loose the ability to identify illegal payers who are not using victims' device though have a genuine payment token.

    34. Quasi-incoherent Physical-Layer Network Coding with FSK Modulation
      Speaker: WANG Zhaorui

      Abstract: Not provided

    35. Adaptive Recoding for BATS Codes
      Speaker: YIN Ho Fai Hoover

      AbstractBATS codes were proposed for communication through networks with packet loss. Most of the existing protocols for BATS codes use an opportunistic recoding approach that transmits a fixed number of recoded packets for all the batches. Intuitively, that is not optimal. A new scheme called adaptive recoding is proposed. The scheme uses a distributed approach which can adapt to random fluctuations in the number of erasures in individual batches and outperform the opportunistic approach. It only requires local knowledge of the received packets and the outgoing network link erasure probability. An algorithm is proposed to solve the scheme efficiently, which can also tolerate rank distribution errors due to inaccurate measurements or limited precision of the machine.

    36. Private Set-Intersection with Access Structure
      Speaker: ZHAO Yongjun

      Abstract: Sharing information to others is common nowadays, but the question is with whom to share. To address this problem, we propose the notion of secret transfer with access structure (STAS). STAS is a two-party computation protocol that enables the server to transfer a secret to a client who satisfies the prescribed access structure. In this paper, we focus on threshold secret transfer (TST), which is STAS for threshold policy and can be made more expressive by using linear secret sharing. TST enables a number of applications including a simple construction of oblivious transfer (OT) with threshold access control, and (a variant of) threshold private set intersection ($t$-PSI), which are the first of their kinds in the literature to the best of our knowledge. The underlying primitive of STAS is a variant of OT, which we call OT for a sparse array. We provide two constructions which are inspired by state-of-the-art PSI techniques including oblivious polynomial evaluation (OPE) and garbled Bloom filter (GBF). The OPE-based construction is secure in the malicious model, while the GBF-based one is more efficient. We implemented the latter one and showed its performance in applications such as privacy-preserving matchmaking.

    37. Cooperative and Competitive Pricing for Mobile Crowdsourced Internet Access
      Speaker: ZHANG Meng

      Abstract: Mobile Crowdsourced Internet Access (MCA) enables mobile users (MUs) to effectively share their Internet connections and tether mobile data to each other, hence improves the overall utilization of network resources. However, MCA can either drive away revenue-generating mobile traffic or introduce unexpected congestions for mobile network operators (MNOs), hence is blocked by some MNOs in practice. In this study, we reconcile the conflicting objectives of MNOs and MUs by designing a novel pricing framework for MCA for the MNOs. We derive the optimal data and tethering pricing schemes systematically in scenarios with operator cooperation and competition. With the widely used $\alpha$-fair MU utility function, we show that the optimal tethering prices are zero and the optimal usage-based data prices are identical for all MUs, in both the cooperative and competitive scenarios. Such optimal schemes also lead to win-win results for the MNOs and MUs. Compared to the case where MCA is blocked, our proposed pricing scheme approximately triples both the MNOs' profit and the MUs' payoff when the MNOs cooperate. Introducing competition among the MNOs will decrease MNOs' profit and further increase the MUs' payoff.

    38. Computationally efficient covert communication
      Speaker: ZHANG Qiaosheng

      Abstract: In this paper, we design the first computationally efficient codes for simultaneously reliable and deniable communication over a Binary Symmetric Channel (BSC). Our setting is as follows – a transmitter Alice wishes to potentially reliably transmit a message to a receiver Bob, while ensuring that the transmission taking place is deniable from eavesdropper Willie (who hears Alice’s transmission over a noisier BSC). Prior works show that Alice can reliably and deniably transmit O(\sqrt(n)) bits over n channel uses without any shared secret between Alice and Bob. One drawback of prior works is that the computational complexity of the codes designed scales as Θ(\sqrt(n)). In this work we provide the first computationally tractable codes with provable guarantees on both reliability and deniability, while simultaneously achieving the best known throughput for the problem.

    39. Optimization of Interference Alignment in MIMO Channel with Multiple Layers of Relays
      Speaker: WEI Yi

      Abstract: Not provided

    40. Delivering Group-based Feedback of Electricity Usage in Campus Dorms
      Speaker: ZHAN Lei

      Abstract: Not provided

    41. An asynchronous load balancing scheme for multi-server systems
      Speaker: LIU Fang

      Abstract: The load balancing problem in asynchronous multi-server systems is considered. Our goal is to find optimal sequence sets which minimize the maximum mean job waiting time for all relative time shifts. We have proved the optimality of SI and DI constructions, and numerically compared their performance with other schemes.

    42. A Mobile Fronthaul System Architecture for Dynamic Provisioning and Protection
      Speaker: YANG Qianmei

      Abstract: This paper presents a privacy-preserving multi-pattern matching system. Privacy: process an encrypted query string over an encrypted pattern set. Application: anti-virus, natural language processing, bioinformatics,etc. A symmetric-key system based on Aho-Corasick automaton.

    43. Privacy preserving multi-pattern matching
      Speaker: WANG Xiuhua

      Abstract: This paper presents a privacy-preserving multi-pattern matching system. Privacy: process an encrypted query string over an encrypted pattern set. Application: anti-virus, natural language processing, bioinformatics,etc. A symmetric-key system based on Aho-Corasick automaton.

    44. Quantum Factor Graphs: Closing-the-Box Operation and Variational Approaches
      Speaker: CAO Xuan

      Abstract: Factor graphs and the sum-product algorithms form a powerful framework for expressing a great variety of algorithms in coding theory, signal processing, artificial intelligence, and other areas. Two different approaches to derive the sum-product algorithm are given by the so-called closing-thebox operation or by a variational approach known as Bethe free energy function minimization. In this paper we consider a generalization of factor graphs known as quantum factor graphs, along with a generalization of the sum-product algorithm known as the quantum sum-product algorithm. We explore the generalization of the closing-the-box operation and the Bethe free energy function from the classical to the quantum setup. Some expressions that hold exactly in the classical case hold only approximately in the quantum case; we give some analytical and numerical characterizations of these approximations.

    45. Low-Complexity Downlink MMSE Beamforming in Massive MIMO Systems
      Speaker: Ye Junhong

      Abstract: Massive MIMO (multiple-input-multiple-output) has been identified as a key technology to improve the cellular system throughput by several orders of magnitude. One practical concern of the implementation of massive MIMO lies in the high computational complexity for coherent and joint signal processing over a large number of antennas. To address this issue, we propose a low-complexity fixed-point algorithm to solve the downlink MMSE beamforming problem in massive MIMO systems. Specifically, the proposed algorithm has a linear complexity in the transmit antenna number, rendering it scalable in large systems. Through both analysis and numerical simulations, we show that the proposed algorithm has much lower complexity than the widely adopted existing algorithms, such as the interior point method and gradient descent method.

    46. Design a Switch-Controller Framework for Data Center Networks
      Speaker: WANG Suzhen

      Abstract: Current data center networks (DCNs) scale out exponentially, Internet traffic patterns become highly dynamic, and network users place more sophisticated demands. As a consequence, traditional networks, where the hardware was coupled with the software, suffer from expansive management, expensive maintain and lack of scalability and flexibility to support more complicated applications. Researchers therefore proposed SDN as a possible solution to handle all these challenging problems. In this poster, we present a controller-switch framework design for DCNs aiming to make deploying and operating in DCNs easier and faster over time.

    47. De novo detection of Differentially Methylated Region
      Speaker: SHEN Linghao

      Abstract: Methylation is a widely studied epigenetic modification. The detection of differentially methylated regions (DMRs) is of great interest for disease studies. We propose a probabilistic model using Hidden Markov Model (HMM) for detection methylation status from methylation array data. We compared our method with existing methods using synthetic data and real data, and our method outperformed other methods.

    48. A Study of Posting and Response Patterns in Tencent's Online Social Network
      Speaker: YING Qiufang

      Abstract: A primary user activity in online social networks (OSN) is to share and receive content posted by friends. What is posted can be broadly divided into two categories: (a) personal content, and (b) public content. An example of personal content is a photo or some text created by the posting person. Public content is usually a link to some online news, or someone else's online blog. The receiving person may respond by either indicating "like" or writing a brief comment. Through these activities, OSN users keep track of what is going on in their "world". We study such social activities in Tencent's social network, in which some of the friend relationships are tagged, for example as "good friend", "school mate", "colleauge", "family" or "business". Studying OSN based on activities rather than only friendship graph let us gain a deeper understanding of social networks. We characterize the types and the distribution of such type of content being shared and diffused. We also explore whether these activity patterns can help predict relationships.

    49. Fading performance of channel coded PNC
      Speaker: Tahernia Mehrdad

      Abstract: Not Provided

    50. Stochastic Process Based Ensemble and Specialized Classifiers
      Speaker: LI Zhizhong

      Abstract: This work explores new methodologies to combine specialized classifiers, i.e., those dedicated to a subset of classes. This is motivated by an empirical study, which reveals the importance of specialized classifiers even with the use of deep architectures. However, we are faced with a key challenge in this endeavor--the restricted scopes of specialized classifiers post serious difficulties to conventional techniques that usually rely on the assumption that all constituents cover the same set of classes. To tackle this difficulty, we develop a novel framework, called Stochastic Process Based Ensemble, which uses stochastic processes to integrate classifiers, and thus derives a coherent perspective over their predictions. This is based on the intuition that the decision process of multiple classifiers can be modeled by a continuous time Markov chain. Both theoretical and experimental study shows that they can work well with a wide range of classifiers, even when their coverage is highly imbalanced. In our experiments, the proposed method resulted in further performance improvement over state-of-the-art deep neural networks on large scale datasets.

    51. Impact of the Uncertainty of Distributed Renewable Generation on Deregulated Electricity Supply Chain
      Speaker: YI Hanling

      Abstract: Microgrids are electricity customers that also produce power to meet their own demand. They are now widely recognized as a clear opportunity towards distributed renewable integration. Despite apparent benefits of incorporating renewable sources in microgrid, uncertainty in renewable generation can impose unprecedented challenges in efficient operation of the existing deregulated electricity supply chain, which is designed to operate with no or little uncertainty in both supply and demand. While most of the previous studies focused on the impact of renewables on the supply side of the supply chain, we investigate the impact of distributed renewable generation on the demand side. In particular, we study how the uncertainty from distributed renewable generation in microgrid affects the average buying cost of utilities and the cost-saving of the microgrid. Our analysis shows that the renewable uncertainty in microgrid can (i) increase the average buying cost of the utility serving the microgrid, termed as local impact, and (ii) somewhat surprisingly, reduce the average buying cost of other utilities participating in the same electricity market, termed as global impact. Moreover, the local impact will lead to an increase in the electricity retail price of microgrid, resulting in a cost-saving less than the case without renewable uncertainty. These observations reveal an inherently economic incentive for utilities to improve their load forecasting accuracy, in order to avoid economy loss and even extract economic benefit in the electricity market. We verify our theoretical results by extensive experiments using real-world traces. Our experimental results show that a 9% increase in load forecasting error (modeled by the standard deviation of the mismatch between real-time actual demand and day-ahead purchased supply) will increase the average buying cost of the utility by 10%.


























.


























Date: November 18, 2016 (Friday)

Time: 2:00pm - 5:00pm

Venue:
Room 603, 6/F,
Ho Sin Hang Engineering Building



Steering Committee:
Awarding Committee:
  • Prof. Chen Lian-Kuan, Prof. Sidharth Jaggi and Prof. ZHANG Kehuan
Organizing Committee:
  • Tongxin LI, Xishi WANG and Qiaosheng ZHANG.
  • Email: ie.poster.day [at] gmail.com

Department of Information Engineering, The Chinese University of Hong Kong.