Digital Object Identifier 10.1109/OJSP.2021.3105305 Editorial Applied Artificial Intelligence and Machine Learning for Video Coding and Streaming We are pleased to present this Special Issue on this timely and impactful topic. In the last few years, hardly a day goes by that we do not hear about the latest advancements and improvements that Artificial Intelligence (AI), and particu- larly its subset Machine Learning (ML) more recently, have brought to a wide spectrum of domains: from technology and medicine to science and sociology, and many others. AI is one of the core enabling components of the fourth industrial revo- lution that we are currently witnessing, and the applications of AI are truly transforming our world and impacting all facets of society, economy, living, working, and technology. The field of Video Technology is no exception and has already been impacted by Applied AI. Video continues to be the dominant traffic on the Internet, and especially due to the COVID-19 pandemic causing increased video usage, video traffic in the USA increased by 70% in 2020 compared to 2019 and consti- tuted 71% of all 2020 IP traffic. 1 Therefore, improving video coding methods and video networking schemes is vital to cope with this increasing demand. In recent years, we have witnessed an exponential growth of applying AI to revolu- tionize the field of video coding and streaming. This is the result of AI and ML having become affordable and practical partially due to accessibility to high processing power, GPUs, and availability of various large datasets. AI and ML-based solutions now offer state-of-the-art in many high-level and low-level image and video related tasks, such as object detection, tracking, segmentation, denoising, filtering, color correction, etc. The power of these tools has recently been introduced to some video coding and video streaming problems as well. A range of Convolutional Neural Network (CNN)-based video coding tools (rate-distortion optimization, deblocking filters, interpolation filters, chroma from luma prediction methods), learned entropy coding, end- to-end image compression techniques, decision tree based encoder speed ups, Fuzzy network bandwidth prediction, and video network resource allocation via reinforcement learning are among these efforts. However, much more effort is required to advance this field, and overcome the existing challenges. Improving the compression efficiency, lowering the overhead of computations of these AI/ML tools, finding suitable loss functions and optimization algorithms, managing network resources according to user experience, accurately predicting network status, and working towards explainable AI are some of these challenges. This Special Issue is dedicated to the said challenges and covers novel applied 1 Dean Takahashi, “Comcast: Pandemic drove peak internet traffic up 32% in 2020”, VentureBeat, March 2, 2021. methods, designs, and systems for AI and ML-based video coding and streaming. Specifically, we invited papers that cover the recent contributions of ML to advances in the whole video processing chain - from creation of high quality video, video compression and flexible representation for optimised distribution, to evaluation of the final Quality of Experience (QoE), all more efficiently achieved thanks to application of ML. It is interesting to note that 7 of the 8 accepted papers either come from the industry or are the result of academia and industry collaboration. We are particularly pleased to see the participation of industry heavyweights such as (in alphabetic order) AVIWEST, BBC, Bitmovin, Ciena, Google, Netflix, and Qualcomm. This is different from the usual academia- dominated special issues of signal processing journals, most likely because of the issue’s focus on applied signal process- ing and ML, and indicates the immediate and real impact that ML is already having in this field. With this in mind, let us now take a look at the summary of the accepted papers and their topics. Reflecting the trend of higher video quality demand, power- ful deep learning methods based on CNNs have recently been successfully deployed in tasks that aim at making pixels better during content production stages. A significant application area of such approaches is enhancement of legacy content which is available in lower qualities despite being profes- sionally produced. Addressing the need for creation of higher frame rate video, the first paper in this special issue proposes “PDWN: Pyramid Deformable Warping Network for Video Interpolation,” Chen et al. [A1]. This approach improves ex- isting methods for frame interpolation in multiple aspects thanks to the coarse-to-fine successive refinement approach with deformable convolutions and feature correlations, which also results in smaller model size and shorter inference time. Interestingly, the model can be extended to use more than the typical two frames, which opens various possibilities for new applications of video enhancement. While new professionally produced content increases user expectations of video quality, the vast amount of complex and diverse User Generated Content (UGC) is created with- out much quality control. However, users are consuming an increasing amount of UGC, creating the need for fast quality prediction of such content. This challenge is addressed in a paper titled “RAPIQUE: Rapid and Accurate Video Quality Prediction of User Generated Content,” Tu et al. [A2]. Be- cause of the diversity of UGC, its quality prediction, even if limited to the spatial domain only, is very complex, while available methodologies for evaluating temporal video aspects This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ 410 VOLUME 2, 2021