Data Quality-Aware Mixed-Precision Quantization via Hybrid Reinforcement Learning.

Mixed-precision quantization mostly predetermines the model bit-width settings before actual training due to the non-differential bit-width sampling process, obtaining suboptimal performance. Worse still, the conventional static quality-consistent training setting, i.e., all data is assumed to be of the same quality across training and inference, overlooks data quality changes in real-world applications which may lead to poor robustness of the quantized models. In this article, we propose a novel data quality-aware mixed-precision quantization framework, dubbed DQMQ, to dynamically adapt quantization bit-widths to different data qualities. The adaption is based on a bit-width decision policy that can be learned jointly with the quantization training. Concretely, DQMQ is modeled as a hybrid reinforcement learning (RL) task that combines model-based policy optimization with supervised quantization training. By relaxing the discrete bit-width sampling to a continuous probability distribution that is encoded with few learnable parameters, DQMQ is differentiable and can be directly optimized end-to-end with a hybrid optimization target considering both task performance and quantization benefits. Trained on mixed-quality image datasets, DQMQ can implicitly select the most proper bit-width for each layer when facing uneven input qualities. Extensive experiments on various benchmark datasets and networks demonstrate the superiority of DQMQ against existing fixed/mixed-precision quantization methods.

Towards performance-maximizing neural network pruning via global channel attention.

Network pruning has attracted increasing attention recently for its capability of transferring large-scale neural networks (e.g., CNNs) into resource-constrained devices. Such a transfer is typically achieved by removing redundant network parameters while retaining its generalization performance in a static or dynamic manner. Concretely, static pruning usually maintains a larger and fit-to-all (samples) compressed network by removing the same channels for all samples, which cannot maximally excavate redundancy in the given network. In contrast, dynamic pruning can adaptively remove (more) different channels for different samples and obtain state-of-the-art performance along with a higher compression ratio. However, since the system has to preserve the complete network information for sample-specific pruning, the dynamic pruning methods are usually not memory-efficient. In this paper, our interest is to explore a static alternative, dubbed GlobalPru, from a different perspective by respecting the differences among data. Specifically, a novel channel attention-based learn-to-rank framework is proposed to learn a global ranking of channels with respect to network redundancy. In this method, each sample-wise (local) channel attention is forced to reach an agreement on the global ranking among different data. Hence, all samples can empirically share the same ranking of channels and make the pruning statically in practice. Extensive experiments on ImageNet, SVHN, and CIFAR-10/100 demonstrate that the proposed GlobalPru achieves superior performance than state-of-the-art static and dynamic pruning methods by significant margins.

Predict-and-Drive: Avatar Motion Adaption in Room-Scale Augmented Reality Telepresence with Heterogeneous Spaces.

Avatar-mediated symmetric Augmented Reality (AR) telepresence has emerged with the ability to empower users located in different remote spaces to interact with each other in 3D through avatars. However, different spaces have heterogeneous structures and features, which bring difficulties in synchronizing avatar motions with real user motions and adapting avatar motions to local scenes. To overcome these issues, existing methods generate mutual movable spaces or retarget the placement of avatars. However, these methods limit the telepresence experience in a small sub-area space, fix the positions of users and avatars, or adjust the beginning/ending positions of avatars without presenting smooth transitions. Moreover, the delay between the avatar retargeting and users' real transitions can break the semantic synchronization between users' verbal conversation and perceived avatar motion. In this paper, we first examine the impact of the aforementioned transition delay and explore the preferred transition style with the existence of such delay through user studies. With the results showing a significant negative effect of avatar transition delay and providing the design choice of the transition style, we propose a Predict-and-Drive controller to diminish the delay and present the smooth transition of the telepresence avatar. We also introduce a grouping component as an upgrade to immediately calculate a coarse virtual target once the user initiates a transition, which could further eliminate the avatar transition delay. Once having the coarse virtual target or an exactly predicted target, we find the corresponding target for the avatar according to the pre-constructed mapping of objects of interest between two spaces. The avatar control component maintains an artificial potential field of the space and drives the avatar towards the target while respecting the obstacles in the physical environment. We further conduct ablation studies to evaluate the effectiveness of our proposed components.

Toward Cost-Effective Mobile Video Streaming through Environment-Aware Watching State Prediction.

Mobile video applications are becoming increasingly prevalent and enriching the way people learn and are entertained. However, on mobile terminals with inherently limited resources, mobile video streaming services consume too much energy and bandwidth, which is an urgent problem to solve. At present, research on cost-effective mobile video streaming typically focuses on the management of data transmission. Among such studies, some new approaches consider the user's behavior to further optimize data transmission. However, these studies have not adequately discussed the specific impact of the physical environment on user behavior. Therefore, this paper takes into account the environment-aware watching state and proposes a cost-effective mobile video streaming scheme to reduce power consumption and mobile data usage. First, the watching state is predicted by machine learning based on user behavior and the physical environment during a given time window. Second, based on the resulting prediction, a downloading algorithm is introduced based on the user equipment (UE) running mode in the LTE system and the VLC player. Finally, according to the corresponding experimental results obtained in a real-world environment, the proposed approach, compared to its benchmarks, effectively reduces the data usage (14.4% lower than that of energy-aware, on average) and power consumption (about 19% when there are screen touches) of mobile devices.

Deploying external bandwidth guaranteed media server clusters for real-time live streaming in media cloud.

The cloud-based media streaming service is a promising paradigm for multimedia applications. It is attractive to media streaming service providers, who wish to deploy their media server clusters in a media cloud at reduced cost. Since the real-time live streaming service is both a bandwidth-intensive and quality-sensitive application, how to optimize the internal bandwidth utilization of a data center network (DCN) as well as guarantee the external bandwidth of the real-time live streaming application, is a key issue of deploying virtual machine (VM)-hosted media server cluster in a media cloud. Therefore, in this study, we propose an external-bandwidth-guaranteed media server cluster deployment scheme in media cloud. The approach simultaneously considers the outside bandwidth requirement of a tree-based media server cluster for live streaming and the intra-bandwidth consumption of a DCN. The proposed scheme models the optimal problem as a new terminal-Steiner-tree-like problem and provides an approximate algorithm for placing the media servers. Our evaluation results show that the proposed scheme guarantees the external bandwidth requirement of a real-time live streaming application, at the same time, greatly reduces the intra-bandwidth utilization of a media cloud with different DCN structures.

Enhancing QoE for Mobile Users by Environment-Aware HTTP Adaptive Streaming.

HTTP adaptive streaming (HAS) has become a dominated media streaming paradigm in today's Internet, which enriches the user's experience by matching the video quality with the dynamic network conditions. A range of HAS mechanisms have been proposed to enhance the Quality of Experience (QoE). However, existing mechanisms ignore the environmental impact in the QoE evaluation of mobile users, while the popularity of mobile video allows users to watch videos in diversified scenarios. In this paper, we propose an environment-aware HAS scheme that fully concentrates on the different criteria for evaluating video QoE under different environments. Using the advantage of the sensors in mobile phones, the scheme constructs and validates a video QoE model based on environment perception and then designs a model-driven, environment-aware HAS rate adaptation algorithm. We also evaluate the scheme with an environment-aware DASH (Dynamic Adaptive Streaming over HTTP) player in real mobile environments. Compared to the benchmark HAS mechanism, the experimental results demonstrate that our scheme can provide appropriate differentiated rate adaptation for different environments, resulting in a higher QoE.

Cluster Cooperation in Wireless-Powered Sensor Networks: Modeling and Performance Analysis.

A wireless-powered sensor network (WPSN) consisting of one hybrid access point (HAP), a near cluster and the corresponding far cluster is investigated in this paper. These sensors are wireless-powered and they transmit information by consuming the harvested energy from signal ejected by the HAP. Sensors are able to harvest energy as well as store the harvested energy. We propose that if sensors in near cluster do not have their own information to transmit, acting as relays, they can help the sensors in a far cluster to forward information to the HAP in an amplify-and-forward (AF) manner. We use a finite Markov chain to model the dynamic variation process of the relay battery, and give a general analyzing model for WPSN with cluster cooperation. Though the model, we deduce the closed-form expression for the outage probability as the metric of this network. Finally, simulation results validate the start point of designing this paper and correctness of theoretical analysis and show how parameters have an effect on system performance. Moreover, it is also known that the outage probability of sensors in far cluster can be drastically reduced without sacrificing the performance of sensors in near cluster if the transmit power of HAP is fairly high. Furthermore, in the aspect of outage performance of far cluster, the proposed scheme significantly outperforms the direct transmission scheme without cooperation.