An overview of the self-organizing map algorithm, on which the papers in this issue are based, is presented in this article.

The Self-Organizing Map

With the emergence of ever-growing advanced vehicular applications, the challenges to meet the demands from both communication and computation are increasingly prominent. Without powerful communication and computational support, various vehicular applications and services will still stay in the concept phase and cannot be put into practice in the daily life. Thus, solving this problem is of great importance. The existing solutions, such as cellular networks, roadside units (RSUs), and mobile cloud computing, are far from perfect because they highly depend on and bear the cost of additional infrastructure deployment. Given tremendous number of vehicles in urban areas, putting these underutilized vehicular resources into use offers great opportunity and value. Therefore, we conceive the idea of utilizing vehicles as the infrastructures for communication and computation, named vehicular fog computing (VFC), which is an architecture that utilizes a collaborative multitude of end-user clients or near-user edge devices to carry out communication and computation, based on better utilization of individual communication and computational resources of each vehicle. By aggregating abundant resources of individual vehicles, the quality of services and applications can be enhanced greatly. In particular, by discussing four types of scenarios of moving and parked vehicles as the communication and computational infrastructures, we carry on a quantitative analysis of the capacities of VFC. We unveil an interesting relationship among the communication capability, connectivity, and mobility of vehicles, and we also find out the characteristics about the pattern of parking behavior, which benefits from the understanding of utilizing the vehicular resources. Finally, we discuss the challenges and open problems in implementing the proposed VFC system as the infrastructures. Our study provides insights for this novel promising paradigm, as well as research topics about vehicular information infrastructures.

Vehicular Fog Computing: A Viewpoint of Vehicles as the Infrastructures

Designed to learn long-range interactions on sequential data, transformers continue to show state-of-the-art results on a wide variety of tasks. In contrast to CNNs, they contain no inductive bias that prioritizes local interactions. This makes them expressive, but also computationally infeasible for long sequences, such as high-resolution images. We demonstrate how combining the effectiveness of the inductive bias of CNNs with the expressivity of transformers enables them to model and thereby synthesize high-resolution images. We show how to (i) use CNNs to learn a context-rich vocabulary of image constituents, and in turn (ii) utilize transformers to efficiently model their composition within high-resolution images. Our approach is readily applied to conditional synthesis tasks, where both non-spatial information, such as object classes, and spatial information, such as segmentations, can control the generated image. In particular, we present the first results on semantically-guided synthesis of megapixel images with transformers and obtain the state of the art among autoregressive models on class-conditional ImageNet. Code and pretrained models can be found at this https URL .

Taming Transformers for High-Resolution Image Synthesis

Radio access technologies for cellular mobile communications are typically characterized by multiple access schemes, e.g., frequency division multiple access (FDMA), time division multiple access (TDMA), code division multiple access (CDMA), and OFDMA. In the 4th generation (4G) mobile communication systems such as Long-Term Evolution (LTE) (Au et al., Uplink contention based SCMA for 5G radio access. Globecom Workshops (GC Wkshps), 2014. doi:10.1109/GLOCOMW.2014.7063547) and LTE-Advanced (Baracca et al., IEEE Trans. Commun., 2011. doi:10.1109/TCOMM.2011.121410.090252; Barry et al., Digital Communication, Kluwer, Dordrecht, 2004), standardized by the 3rd Generation Partnership Project (3GPP), orthogonal multiple access based on OFDMA or single carrier (SC)-FDMA is adopted. Orthogonal multiple access was a reasonable choice for achieving good system-level throughput performance with simple single-user detection. However, considering the trend in 5G, achieving significant gains in capacity and system throughput performance is a high priority requirement in view of the recent exponential increase in the volume of mobile traffic. In addition the proposed system should be able to support enhanced delay-sensitive high-volume services such as video streaming and cloud computing. Another high-level target of 5G is reduced cost, higher energy efficiency and robustness against emergencies.

Non-Orthogonal Multiple Access (NOMA) for cellular future radio access

Several recent work on speech synthesis have employed generative adversarial networks (GANs) to produce raw waveforms. Although such methods improve the sampling efficiency and memory usage, their sample quality has not yet reached that of autoregressive and flow-based generative models. In this work, we propose HiFi-GAN, which achieves both efficient and high-fidelity speech synthesis. As speech audio consists of sinusoidal signals with various periods, we demonstrate that modeling periodic patterns of an audio is crucial for enhancing sample quality. A subjective human evaluation (mean opinion score, MOS) of a single speaker dataset indicates that our proposed method demonstrates similarity to human quality while generating 22.05 kHz high-fidelity audio 167.9 times faster than real-time on a single V100 GPU. We further show the generality of HiFi-GAN to the mel-spectrogram inversion of unseen speakers and end-to-end speech synthesis. Finally, a small footprint version of HiFi-GAN generates samples 13.4 times faster than real-time on CPU with comparable quality to an autoregressive counterpart.

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis

Although end-to-end neural text-to-speech (TTS) methods (such as Tacotron2) are proposed and achieve state-of-theart performance, they still suffer from two problems: 1) low efficiency during training and inference; 2) hard to model long dependency using current recurrent neural networks (RNNs). Inspired by the success of Transformer network in neural machine translation (NMT), in this paper, we introduce and adapt the multi-head attention mechanism to replace the RNN structures and also the original attention mechanism in Tacotron2. With the help of multi-head self-attention, the hidden states in the encoder and decoder are constructed in parallel, which improves training efficiency. Meanwhile, any two inputs at different times are connected directly by a self-attention mechanism, which solves the long range dependency problem effectively. Using phoneme sequences as input, our Transformer TTS network generates mel spectrograms, followed by a WaveNet vocoder to output the final audio results. Experiments are conducted to test the efficiency and performance of our new network. For the efficiency, our Transformer TTS network can speed up the training about 4.25 times faster compared with Tacotron2. For the performance, rigorous human tests show that our proposed model achieves state-of-the-art performance (outperforms Tacotron2 with a gap of 0.048) and is very close to human quality (4.39 vs 4.44 in MOS).

/pdf/neural-speech-synthesis-with-transformer-network-4xysk2sb4r.pdf

Neural Speech Synthesis with Transformer Network.

The most important issue that must be solved in designing a data gathering algorithm for wireless sensor networks (WSNS) is how to save sensor node energy while meeting the needs of applications/users. In this paper, we propose a novel energy-aware routing protocol (EAP) for a long-lived sensor network. EAP achieves a good performance in terms of lifetime by minimizing energy consumption for in-network communications and balancing the energy load among all the nodes. EAP introduces a new clustering parameter for cluster head election, which can better handle the heterogeneous energy capacities. Furthermore, it also introduces a simple but efficient approach, namely, intra-cluster coverage to cope with the area coverage problem. We use a simple temperature sensing application to evaluate the performance of EAP and results show that our protocol significantly outperforms LEACH and HEED in terms of network lifetime and the amount of data gathered.

https://www.mdpi.com/1424-8220/9/1/445/pdf

An energy-aware routing protocol in wireless sensor networks

Although end-to-end neural text-to-speech (TTS) methods (such as Tacotron2) are proposed and achieve state-of-the-art performance, they still suffer from two problems: 1) low efficiency during training and inference; 2) hard to model long dependency using current recurrent neural networks (RNNs). Inspired by the success of Transformer network in neural machine translation (NMT), in this paper, we introduce and adapt the multi-head attention mechanism to replace the RNN structures and also the original attention mechanism in Tacotron2. With the help of multi-head self-attention, the hidden states in the encoder and decoder are constructed in parallel, which improves the training efficiency. Meanwhile, any two inputs at different times are connected directly by self-attention mechanism, which solves the long range dependency problem effectively. Using phoneme sequences as input, our Transformer TTS network generates mel spectrograms, followed by a WaveNet vocoder to output the final audio results. Experiments are conducted to test the efficiency and performance of our new network. For the efficiency, our Transformer TTS network can speed up the training about 4.25 times faster compared with Tacotron2. For the performance, rigorous human tests show that our proposed model achieves state-of-the-art performance (outperforms Tacotron2 with a gap of 0.048) and is very close to human quality (4.39 vs 4.44 in MOS).

Neural Speech Synthesis with Transformer Network

Maximizing network lifetime based on transmission range adjustment in wireless sensor networks

In Vehicular Ad Hoc Networks (VANETs), the major communication challenge lies in very poor connectivity, which can be caused by sparse or unbalanced traffic. Deploying supporting infrastructure could relieve this problem, but it often requires a large amount of investment and elaborate design, especially at the city scale. In this paper, we propose the idea of Parked Vehicle Assistance (PVA), which allows parked vehicles to join VANETs as static nodes. With wireless device and rechargable battery, parked vehicles can easily communicate with one another and their moving counterparts. Owing to the extensive parking in cities, parked vehicles are natural roadside nodes characterized by large number, long-time staying, wide distribution, and specific location. So parked vehicles can serve as static backbone and service infrastructure to improve connectivity. We investigate network connectivity in PVA through theoretic analysis and realistic survey and simulations. The results prove that even a small proportion of PVA vehicles could overcome sparse or unbalanced traffic, and promote network connectivity greatly. Thus, PVA enhances VANETs from down to top, and paves the way for new hybrid networks with static and mobile nodes.

/pdf/pva-in-vanets-stopped-cars-are-not-silent-26brg3abam.pdf

Ming Liu

Papers

Neural Speech Synthesis with Transformer Network.

An energy-aware routing protocol in wireless sensor networks

Neural Speech Synthesis with Transformer Network

Maximizing network lifetime based on transmission range adjustment in wireless sensor networks

PVA in VANETs: Stopped cars are not silent