Towards Low Latency Multi-viewpoint 360° Interactive Video: A Multimodal Deep Reinforcement Learning Approach

Pang, Haitian; Zhang, Cong; Wang, Fangxin; Liu, Jiangchuan; Sun, Lifeng

doi:10.1109/infocom.2019.8737395

Cited by 28 publications

(10 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Similarly, also Deep Reinforcement Learning (DRL) has been successfully used in a plethora of different contexts in the networking field. Among others, we mention resource allocation and network virtualization [28,29,[40][41][42], dynamic spectrum access [30,43,44], cellular networks [45,46], rate selection [31], and multimedia streaming [32][33][34][35][36]. Although DRL has been frequently used for optimization of networking metrics, it has seldom been used so far to improve the performance of other learning systems in the networking context.…”

Section: Related Workmentioning

confidence: 99%

“…This paper presents Chares, a real-time and model-free Deep Reinforcement Learning (DRL) approach that leverages waveform synthesis to improve the resilience of WSC applications via channel-specific FIR filtering (Section III). Beyond playing video games [27], DRL has experienced a surge of interest in the wireless networking community as well [28][29][30][31][32][33][34][35][36]. This is because DRL provides a very general framework based on partially-observable Markov decision process (POMDP), which allows to dynamically solve a multitude of problems without explicit modeling.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Can You Fix My Neural Network? Real-Time Adaptive Waveform Synthesis for Resilient Wireless Signal Classification

Restuccia

Melodia

2021

IEEE INFOCOM 2021 - IEEE Conference on Computer Communications

View full text Add to dashboard Cite

Due to the sheer scale of the Internet of Things (IoT) and 5G, the wireless spectrum is becoming severely congested. For this reason, wireless devices will need to continuously adapt to current spectrum conditions by changing their communication parameters in real-time. Therefore, wireless signal classification (WSC) will become a compelling necessity to decode fast-changing signals from dynamic transmitters. Thanks to its capability of classifying complex phenomena without explicit mathematical modeling, deep learning (DL) has been demonstrated to be a key enabler of WSC. Although DL can achieve a very high accuracy under certain conditions, recent research has unveiled that the wireless channel can disrupt the features learned by the DL model during training, thus drastically reducing the classification performance in real-world live settings. Since retraining classifiers is cumbersome after deployment, existing work has leveraged the usage of carefully-tailored Finite Impulse Response (FIR) filters that, when applied at the transmitter's side, can restore the features that are lost because of the the channel actions, i.e., waveform synthesis. However, these approaches compute FIRs using offline optimization strategies, which limits their efficacy in highly-dynamic channel settings. In this paper, we improve the state of the art by proposing Chares, a Deep Reinforcement Learning (DRL)-based framework for channel-resilient adaptive waveform synthesis. Chares adapts to new and unseen channel conditions by optimally computing through DRL the FIRs in real time. Chares is a DRL agent whose architecture is based upon the Twin Delayed Deep Deterministic Policy Gradients (TD3), which requires minimal feedback from the receiver and explores a continuous action space for best performance. Chares has been extensively evaluated on two well-known datasets with an extensive number of channels. We have also evaluated the real-time latency of Chares with an implementation on field-programmable gate array (FPGA). Results show that Chares increases the accuracy up to 4.1x when no waveform synthesis is performed, by 1.9x with respect to existing work, and can compute new actions within 41µs.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Can You Fix My Neural Network? Real-Time Adaptive Waveform Synthesis for Resilient Wireless Signal Classification

Restuccia

Melodia

2021

IEEE INFOCOM 2021 - IEEE Conference on Computer Communications

View full text Add to dashboard Cite

show abstract

“…Zhang et al in [15] proposed a deep reinforcement learning (DRL) algorithm to learn the user's FoV and bandwidth. Pang in [16] proposed a DRL algorithm to learn the user's FoV and head movements. The learning in DRL algorithms requires random exploration.…”

Section: Related Workmentioning

confidence: 99%

“…To alleviate the impact of the variation of the downloading capacity, we modify Algorithm 1 by restricting the increase or decrease of the bitrate of each tile to be at most one level at each time. Specifically, after computing r o i for segment i ∈ I according to (16)…”

Section: Algorithm Modificationmentioning

confidence: 99%

Online Bitrate Selection for Viewport Adaptive 360-Degree Video Streaming

Tang

Wong

2020

Preprint

View full text Add to dashboard Cite

360-degree video streaming provides users with immersive experience by letting users determine their field-of-views (FoVs) in real time. To enhance the users' quality of experience (QoE) given their limited bandwidth, recent works have proposed a viewport adaptive 360-degree video streaming model by exploiting the bitrate adaptation in spatial and temporal domains. Under this video streaming model, in this paper, we consider a scenario with a newly generated 360-degree video without viewing history from other users. To maximize the user's QoE, we propose an online bitrate selection algorithm, called OBS360. The proposed online algorithm can adapt to the unknown and heterogeneous users' FoVs and downloading capacities. We prove that the proposed algorithm achieves sublinear dynamic regret under a convex decision set. This suggests that as the number of video segments increases, the performance of the online algorithm approaches the performance of the offline algorithm, where the users' FoVs and downloading capacities are known. We perform simulations with real-world dataset to evaluate the performance of the proposed algorithm. Results show that compared with several existing methods, our proposed algorithm can enhance the users' QoE significantly by improving the viewing bitrate and reducing the inter-segment and intra-segment degradation losses of the users.

show abstract

“…For example, in AR [2], [3], [15], video object classification and recognition task has to be performed first and then the videos are delivered to the user. In multi-viewpoint 360 degree interactive video transmission, the viewing-related features have to be analyzed at the edge first, then the video quality and other video transmission related parameters will be determined [16]. Future communication networks will require not only wireless content caching, but also considerable content processing at the edge.…”

Section: Introductionmentioning

confidence: 99%

Joint Video Caching and Processing for Multi-Bitrate Videos in Ultra-Dense HetNets

Zhang

Mao

2020

IEEE Open J. Commun. Soc.

View full text Add to dashboard Cite

Caching popular videos at the edge has been confirmed as a promising way to support low-latency video transmission and alleviate the backhaul traffic burden. Meanwhile, mobile edge computing (MEC) has also been regarded as an effective solution to meet the 5G low-latency service requirements. In this paper, we propose to fully utilize both the storage and computing resources at edge servers to support multiple bitrate video streaming. We design the video caching, processing, and user association models that aim to minimize the average retrieval latency of all users. This problem is modeled as a mixed-integer bilinear problem, which is NP-hard. We show that under practical constraints on storage, bandwidth, and processing capacity, the problem does not exhibit sub-modular property and the performance of a greedy algorithm may not be strictly guaranteed. To deal with this challenging problem, we decompose the original problem into a cache placement problem and a user-BS association problem, while still preserving the interplay between the two sub-problems. A linearization and rounding algorithm, including: (i) a greedy rounding proactively caching scheme and (ii) a random-rounding user-BS association scheme, is then proposed, with performance bounds derived. Extensive simulation results show that the proposed scheme can achieve a near-optimal performance under various storage, computing capacity, and downlink bandwidth settings.Index Terms-Video caching, video transcoding, multiple bitrate video, mobile edge computing (MEC), submodularity.

show abstract

Towards Low Latency Multi-viewpoint 360° Interactive Video: A Multimodal Deep Reinforcement Learning Approach

Cited by 28 publications

References 24 publications

Can You Fix My Neural Network? Real-Time Adaptive Waveform Synthesis for Resilient Wireless Signal Classification

Can You Fix My Neural Network? Real-Time Adaptive Waveform Synthesis for Resilient Wireless Signal Classification

Online Bitrate Selection for Viewport Adaptive 360-Degree Video Streaming

Joint Video Caching and Processing for Multi-Bitrate Videos in Ultra-Dense HetNets

Contact Info

Product

Resources

About