The development of large language models (LLMs) poses new challenges in data center network topology design. To assist in exploring topology design, we propose ATOP, an Automated Topology Optimization Pipeline, which models network topology as a set of hyperparameters, enabling the discovery of potential topologies. With various optimization algorithms and customizable optimization objectives, ATOP achieves automated topology optimization on a scale of tens of thousands of GPUs. We apply ATOP on network topologies for 256, 1024, 4096, and 16384 GPUs, optimizing performance under LLMs training traffic patterns, collective communication performance, fault tolerance, and network cost. We also evaluate ATOP in different scenarios: building, optimizing, and expanding a data center. From ATOP's results, we discover a new topology — ZCube, which reaches the highest cost-effectiveness across various GPU scales. Simulation results show that ZCube, compared to the previous state-of-the-art topologies, including Rail-optimized Fat-tree (ROFT), Rail-only, and HPN, improves end-to-end LLM training speed by 3% to 7% and reduces network hardware costs by 26% to 46%. We also construct ZCube on a real-world testbed. Results show that ZCube reduces hardware costs by 25% compared to Rail-Optimized Topology while maintaining the same all-reduce and all-to-all performance.

Mobile devices such as smartphones, laptops, and tablets can often connect to multiple access networks (e.g., Wi-Fi, LTE, and 5G) simultaneously.Recent advancements facilitate seamless integration of these connections below the transport layer, enhancing the experience for apps that lack inherent multi-path support.This optimization hinges on dynamically determining the traffic distribution across networks for each device, a process referred to as multi-access traffic splitting.This paper introduces NetworkGym, a high-fidelity network environment simulator that facilitates generating multiple network traffic flows and multi-access traffic splitting.This simulator facilitates training and evaluating different RL-based solutions for the multi-access traffic splitting problem.Our initial explorations demonstrate that the majority of existing state-of-the-art offline RL algorithms (e.g. CQL) fail to outperform certain hand-crafted heuristic policies on average.This illustrates the urgent need to evaluate offline RL algorithms against a broader range of benchmarks, rather than relying solely on popular ones such as D4RL.We also propose an extension to the TD3+BC algorithm, named Pessimistic TD3 (PTD3), and demonstrate that it outperforms many state-of-the-art offline RL algorithms.PTD3's behavioral constraint mechanism, which relies on value-function pessimism, is theoretically motivated and relatively simple to implement.We open source our code and offline datasets at github.com/hmomin/networkgym.

Single Image Super-Resolution (SISR) is essential for many computer vision tasks. In some real-world applications, such as object recognition and image classification, the captured image size can be arbitrary while the required image size is fixed, which necessitates SISR with arbitrary scaling factors. It is a challenging problem to take a single model to accomplish the SISR task under arbitrary scaling factors. To solve that problem, this paper proposes a bilateral upsampling network which consists of a bilateral upsampling filter and a depthwise feature upsampling convolutional layer. The bilateral upsampling filter is made up of two upsampling filters, including a spatial upsampling filter and a range upsampling filter. With the introduction of the range upsampling filter, the weights of the bilateral upsampling filter can be adaptively learned under different scaling factors and different pixel values. The output of the bilateral upsampling filter is then provided to the depthwise feature upsampling convolutional layer, which upsamples the low-resolution (LR) feature map to the high-resolution (HR) feature space depthwisely and well recovers the structural information of the HR feature map. The depthwise feature upsampling convolutional layer can not only efficiently reduce the computational cost of the weight prediction of the bilateral upsampling filter, but also accurately recover the textual details of the HR feature map. Experiments on benchmark datasets demonstrate that the proposed bilateral upsampling network can achieve better performance than some state-of-the-art SISR methods.

For many face-related multimedia applications, low-resolution face images may greatly degrade the face recognition performance and necessitate face super-resolution (SR). Among the current SR methods, MSE-oriented SR methods often produce over-smoothed outputs and could miss some texture details while GAN-oriented SR methods may generate artifacts which are harmful to face recognition. To resolve the above issues, this paper presents a supervised pixel-wise Generative Adversarial Network (SPGAN) that can resolve a very low-resolution face image of$16\times 16$or smaller pixel-size to its larger version of multiple scaling factors ($2\times$,$4\times$,$8\times$and even$16\times$) in a unified framework. Being different from traditional unsupervised discriminators which generate a single number to represent the likelihood whether the input image is real or fake, the proposed supervised pixel-wise discriminator mainly focus on whether each pixel of the generated SR face image is as photo-realistic as its corresponding pixel in the ground-truth HR (high-resolution) face image. To further improve the face recognition performance of SPGAN, we take advantage of the face identity prior by sending two inputs to the discriminator, including an input face image (either a real HR face image or its corresponding SR face image) and its face features which are extracted from a pre-trained face recognition model. Due to the introduced face identity prior, the identity-based discriminator can pay more attention to texture details which are closely related to face recognition. Extensive experiments demonstrate that the proposed SPGAN can achieve more photo-realistic SR images and higher face recognition accuracy than some state-of-the-art methods.

Cloud computing technologies can not satisfy the requirements of applications on the mobile terminals because of their disadvantages in delay, link load and energy. So Mobile Edge Computing (MEC) is proposed as a kind of novel computing technology. As an important research direction of MEC, service migration methods still have limitations that they cannot learn migration paths and be adaptive in dynamic situation and user movement. In this paper, we propose a novel service migration policy method based on reinforcement learning. We firstly investigate user movement, four different edge network situations and traditional migration policies. Then we formulate the system requirements by Satisfiability Modulo Theory (SMT) logic to acquire the migration policy space. We further propose a dynamic-awareness deep Q-learning algorithm to select paths from the policy space iteratively and conduct dynamic awareness to adjust learning rate adaptively. Meanwhile, the optimal convergence of our algorithm is proved theoretically. Finally, the experimental results highlight the effectiveness as migration successful rate, service interruption time and load balance of our method compared to the other solutions.

The advent of the next iteration of mobile and wireless communication standards, the so called 5G, is already a reality. 3GPP released in December 2017 the first set of specifications of the 5G New Radio (NR), which introduced important innovations with respect to legacy networks. One of the main novelties is the use of very-high frequencies in the radio access, which requires highly-directional transmissions or beams to overcome the severe propagation losses. Therefore, it is paramount to manage these beams in an efficient manner in order to always choose the optimum set of beams. In this work, we describe the first NR-compliant beam management framework for the ns-3 network simulator. We aim at providing an open-source and fully-customizable solution to let the scientific community implement their solutions and assess their impact on the end-to-end network performance. Additionally, we describe the necessary modifications in ns-3 to align the radio frame structure to what the 3GPP standards mandate. Finally, we validate our results by running a simple mobility scenario.

The millimeter-wave (mmWave) frequencies have attracted considerable attention for fifth generation (5G) cellular communication as they offer orders of magnitude greater bandwidth than current systems. However, the medium access control (MAC) layer may need to be significantly redesigned to support the highly directional transmissions, and the demand for ultra-low latencies and high peak rates expected in mmWave communication. To address these challenges, we present a novel mmWave MAC layer frame structure with a number of enhancements, including flexible, highly granular transmission times, dynamic control signal locations, extended messaging, and the ability to efficiently multiplex directional control signals. Analytic formulas are derived for the utilization and control overhead as a function of control periodicity, number of users, traffic statistics, signal-to-noise ratio, and antenna gains. Importantly, the analysis can incorporate various front-end MIMO capability assumptions-a critical feature of mmWave. Under realistic system and traffic assumptions, the analysis reveals that the proposed flexible frame structure design offers significant benefits over designs with fixed frame structures similar to current 4G long-term evolution. It is also shown that the fully digital beamforming architectures offer significantly lower overhead compared with analog and hybrid beamforming under equivalent power budgets.

Comparing to the low level local features, Transform Invariant Low-Rank Textures (TILT) can in some sense globally rectify a large class of low-rank textures in 2D images, and thus more accurate and robust. However, TILT is still rather rudimentary, and have some limitations in applications. In this paper, we proposed a novel algorithm for better solving TILT. Our method is based on the application of Bayesian framework in robust principal component analysis (RPCA), besides less local minima, nonparametric Bayesian method introduces the uncertainty in the parameters, which make our new algorithm can handle more complex situations. Experimental results on both synthetic and real data indicate that our new algorithm outperforms the existing algorithm especially for the case with corruptions and occlusions.

The increasing demand of data, along with the spectrum scarcity, are motivating a urgent shift towards exploiting new bands. This is the main reason behind identifying mmWaves as the key disruptive enabling technology for 5G cellular networks. Indeed, utilizing new bands means facing new challenges; in this context, they are mainly related to the radio propagation, which is shorter in range and more sensitive to obstacles. The resulting key aspects that need to be taken into account when designing mmWave cellular systems are directionality and link intermittency. The lack of network level results motivated this work, which aims at providing the first of a kind open source mmWave framework, based on the network simulator ns-3. The main focus of this work is the modeling of customizable channel, physical (PHY) and medium access control (MAC) layers for mmWave systems. The overall design and architecture of the model are discussed in details. Finally, the validity of our proposed framework is corroborated through the simulation of a simple scenario.