Making sense of the surrounding context and ongoing events through not only the visual inputs but also acoustic cues is critical for various AI applications. This paper presents an attempt to learn a neural network model that recognizes more than 500 different sound events from the audio part of user generated videos (UGV). Aside from the large number of categories and the diverse recording conditions found in UGV, the task is challenging because a sound event may occur only for a short period of time in a video clip. Our model specifically tackles this issue by combining a main subnet that aggregates information from the entire clip to make clip-level predictions, and a supplementary subnet that examines each short segment of the clip for segment-level predictions. As the labeled data available for model training are typically on the clip level, the latter subnet learns to pay attention to segments selectively to facilitate attentional segmentlevel supervision. We call our model the M&mnet, for it leverages both “M”acro (clip-level) supervision and “m”icro (segment-level) supervision derived from the macro one. Our experiments show that M&mnet works remarkably well for recognizing sound events, establishing a new state-of-theart for DCASE17 and AudioSet data sets. Qualitative analysis suggests that our model exhibits strong gains for short events. In addition, we show that the micro subnet is computationally light and we can use multiple micro subnets to better exploit information in different temporal scales.

---

With the large expected demand of wireless communication, Device-to-Device (D2D) communication has been proposed as a promising technology to enhance network performance. Nevertheless, the selfish nature of potential D2D users may impale the performance of D2D-enabled network. In this paper, we propose a D2D-enabled cellular network framework, which support a novel group D2D mode under overlay D2D communication. The group-based design is derived from the discussions of two common D2D modes, divided and shared D2D modes, regarded as special cases. The proposed framework provides a pricing-based dynamic Stackelberg game for optimal mode selection and spectrum partitioning. We propose the incentive compatible pricing strategy to provide proper incentive for these selfish potential D2D pairs to make optimal choices in mode selection. Our results show that the pricing and spectrum partition strategy effectively prevents selfish potential D2D users from harming the system performance while fully exploits the potential of D2D communication.

---

This letter investigates the source and relay precoder design for full-duplex multiple-input multiple-output relay systems, where simultaneous wireless information and power transfer is enabled at the destination. The objective is to design the precoders such that the end-to-end performance can be optimized. Different from existing schemes, a novel dual-objective function is adopted in this work. The proposed precoders yield closed-form solutions and avoid iterative algorithms. Moreover, our design is applicable when the system suffers from the residual loop-interference. Simulations show that the proposed scheme enables an efcient way to optimize information-decoding and energy-harvesting performances.

---

Mobile devices running augmented reality applications consume considerable energy for graphics-intensive workloads. This paper presents a scheme for the differentiated handling of camera-captured physical scenes and computer-generated virtual objects according to different perceptual quality metrics.We propose online algorithms and their realtime implementations to reduce energy consumption through dynamic frame rate adaptation while maintaining the visual quality required for augmented reality applications. To evaluate system efficacy, we integrate our scheme into Android and conduct extensive experiments on a commercial smartphone with various application scenarios. The results show that the proposed scheme can achieve energy savings of up to 39.1% in comparison to the native graphics system in Android while maintaining satisfactory visual quality.

---

In this paper, we address co-saliency detection in a set of images jointly covering objects of a specic class by an unsupervised convolutional neural network (CNN). Our method does not require any additional training data in the form of object masks. We decompose co-saliency detection into two sub-tasks, single-image saliency detection and cross-image co-occurrence region discovery corresponding to two novel unsupervised losses, the single-image saliency (SIS) loss and the co-occurrence (COOC) loss. The two losses are modeled on a graphical model where the former and the latter act as the unary and pairwise terms, respectively. These two tasks can be jointly optimized for generating co-saliency maps of high quality. Furthermore, the quality of the generated co-saliency maps can be enhanced via two extensions: map sharpening by self-paced learning and boundary preserving by fully connected conditional random elds. Experiments show that our method achieves superior results, even outperforming many supervised methods.

---

We consider rescue missions in postdisaster scenarios with particularly difficult environments where no infrastructure is available. Given the increasing popularity of smartphones and wearable devices, this paper proposes a rescue system which uses the Doppler effect to determine the direction of Wi-Fi signals emitted from disaster survivors' mobile devices to help rescuers quickly locate the survivors. First, we investigate the impact of the search and rescue environment on the direction-finding accuracy of Doppler effect to identify the major challenge and several implementation issues of the system. Then, to address the major challenge of Doppler shifts being too small, we propose an algorithm, which consists of three mechanisms, to solve the problem with the objective of maximizing the direction-finding accuracy. These mechanisms improve the direction-finding accuracy via eliminating the frequency fluctuation as much as possible and improving the sensitivity on small frequency shifts. Also, an active detection scheme is proposed to ensure that the survivors' devices emit steady and continuous Wi-Fi signals, along with a decision logic to minimize energy consumption by the active scheme. We implement the rescue system as a mobile application on Android smartphones and conduct extensive experiments in real-world environments. Results show that the proposed system can reduce rescue times by up to half while consuming reasonable amounts of energy from survivor smartphones.

---

Mobile social networks (MSNs) enable users to discover and share contents with each other, especially at ephemeral events such as exhibitions and conferences where users could be strangers. Nevertheless, the incentive of users to actively share their contents in MSNs may be lacking if the corresponding cost is high. Besides, as users in MSN share contents in an impromptu way as they move, it makes them vulnerable to malicious users who may want to disseminate false contents. This is because users may not have knowledge about the peers they are socially connecting with in the network. In this paper, we propose MCoST, a mechanism that motivates content sharing in MSN and ensures that only trustworthy contents are shared. The mechanism is built on users' collective bidding, content cost sharing, and trust evaluation while guaranteeing individual rationality. MCoST enables content providers to share contents with multiple users simultaneously by utilizing the broadcast nature of wireless transmission. The cost of the content is collectively compensated by the content receivers through the content bidding mechanism in MCoST. In ensuring that users can establish the trustworthiness of their encounters' contents, MCoST incorporates a robust trust evaluation framework that guarantees that content reviews are immutable and tamper-proof, resistive to sybil, and rejection attacks, and that users cannot have multiple and fake identities in the network or reject negative reviews about their contents. This is achieved by integrating a distributed cryptographic hash-chained content review mechanism in the design of MCoST. Performance evaluation shows that the proposed mechanism efficiently evaluates contents' trustworthiness by detecting and discriminating review-chains under sybil or rejection attacks and reduces the time and cost to collect the desired contents by 86% and 40%, respectively, and improves network utilization by 50%.

---

Object co-segmentation aims to segment the common objects in images. This paper presents a CNN-based method that is unsupervised and end-to-end trainable to better solve this task. Our method is unsupervised in the sense that it does not require any training data in the form of object masks but merely a set of images jointly covering objects of a specific class. Our method comprises two collaborative CNN modules, a feature extractor, and a co-attention map generator. The former module extracts the features of the estimated objects and backgrounds, and is derived based on the proposed co-attention loss, which minimizes inter-image object discrepancy while maximizing intra-image figure-ground separation. The latter module is learned to generate co-attention maps by which the estimated figure-ground segmentation can better fit the former module. Besides the co-attention loss, the mask loss is developed to retain the whole objects and remove noises. Experiments show that our method achieves superior results, even outperforming the state-of-the-art, supervised methods.

---

Self-powered intermittent systems featuring nonvolatile processors (NVPs) allow for accumulative execution in unstable power environments. However, frequent power failures may cause incorrect NVP execution results due to invalid data generated intermittently. This paper presents a HW/SW co-design, called HomeRun, to guarantee atomicity by ensuring that an uninterruptible program section can be run through at one execution. We design a HW module to ensure that a power pulse is sufficient for an atomic section, and develop a SW mechanism for programmers to protect atomic sections. The proposed design is validated through the development of a prototype pattern locking system. Experimental results demonstrate that the proposed design can completely guarantee atomicity and significantly improve the energy utilization of self-powered intermittent systems.

---

The increasingly high display resolution of mobile devices imposes a further burden on energy consumption. Existing schemes manage either OLED or GPU power to save energy. This paper presents the design, algorithm, and implementation of a co-managing scheme called Duet, which automatically trades off perceptual quality for energy efficiency in accordance with static and dynamic visual acuity when users interact with mobile applications. The results of experiments conducted on a commercial smartphone with popular interactive apps show that Duet saves more energy while retaining better visual quality, compared with a joint scheme that simultaneously uses dynamic pixel disabling and dynamic resolution scaling to save OLED and GPU energy in isolation.

---