Video super-resolution is one of the most popular tasks on mobile devices, being widely used for an automatic improvement of lowbitrate and low-resolution video streams. While numerous solutions have been proposed for this problem, they are usually quite computationally demanding, demonstrating low FPS rates and power efficiency on mobile devices. In this Mobile AI challenge, we address this problem and propose the participants to design an end-to-end real-time video super-resolution solution for mobile NPUs optimized for low energy consumption. The participants were provided with the REDS training dataset containing video sequences for a 4X video upscaling task. The runtime and power efficiency of all models was evaluated on the powerful MediaTek Dimensity 9000 platform with a dedicated AI processing unit capable of accelerating floating-point and quantized neural networks. All proposed solutions are fully compatible with the above NPU, demonstrating an up to 500 FPS rate and 0.2 [Watt / 30 FPS] power consumption. A detailed description of all models developed in the challenge is provided in this paper.

No abstract

Network intrusion detection system plays a crucial role in protecting the integrity and availability of sensitive assets. Network traffic data contains a large amount of time, space, and statistical information. Existing research lacks the utilization of spatial-temporal multi-granularity data features and the mutual support among different data features, thus making it difficult to specifically and accurately identify anomalies. Taking into account the distinctions among different granularities, we propose a framework called Tri-Broad Learning System (TBLS), which can learn and integrate the three granular features. In order to accurately explore the spatial-temporal connotation of the traffic information, a feature dataset containing three granularities is constructed according to the characteristics of time, space, and data content. In this way, we use broad learning basic units to extract abstract features of different granularities, and then express these features in different feature spaces to enhance them separately. We use the "He" instead of the original initialization method in BLS to initialize the weights of feature nodes and enhancement nodes to achieve better detection accuracy. We exhibit the merits of our proposed model on the UNSW-NB15, CIC-IDS-2017, and mixed traffic datasets. Experimental JOURNAL OF SUPERCOMPUTING results show that TBLS outperforms the typical BLS in terms of various evaluation metrics and time consumption. Compared with other machine learning methods, TBLS achieves better performance metrics.

Wind Power calculation is a key work in wind farm construction. At present, most company use WASP to evaluate the wind power. WASP is designed on the basis of European landform. In china, the landform is so complex that it is need to take a new calculation way to get a more practical result. Here weibull density function is regarded as wind frequency distribution and a simple generation calculation formula is deducted. Based on Chinese complex landform, a constraint condition is added to the calculation expression and a new feasible calculation model is given. VC++ and matlab are used to implement this calculation model. At last, Nordex N80 and N90 are took as an example to evaluate the wind power of an island. The result shows that the performance is better than WASP.

No abstract