While self-attention mechanism has shown promising results for many vision tasks, it only considers the current features at a time. We show that such a manner cannot take full advantage of the attention mechanism. In this paper, we present Deep Connected Attention Network (DCANet), a novel design that boosts attention modules in a CNN model without any modification of the internal structure. To achieve this, we interconnect adjacent attention blocks, making information flow among attention blocks possible. With DCANet, all attention blocks in a CNN model are trained jointly, which improves the ability of attention learning. Our DCANet is generic. It is not limited to a specific attention module or base network architecture. Experimental results on ImageNet and MS COCO benchmarks show that DCANet consistently outperforms the state-of-the-art attention modules with a minimal additional computational overhead in all test cases. The code is available at: https://github.com/13952522076/DCANet.

In this paper, we present a conceptually simple, yet powerful method for image recognition. The method, called Cooperative Dynamic Convolution (CoConv), introduces a cooperative learning of dynamic convolution from multiple convolutional experts. CoConv can be used as a substitute for the traditional static convolution, and can be seamlessly integrated in various visual models. Moreover, CoConv is easy to train with only a minimal computational overhead introduced in the inference phase. CoConv is trained by using multiple convolutional experts simultaneously, and the convolutional weights are merged by a weighted summation before convolutional operations for efficiency during inference. Results from extensive experiments show that CoConv leads to consistent improvement for image classification on various datasets, independent of the choice of the base convolutional network. Remarkably, CoConv improves the top-1 classification accuracy of ResNet18 by 3.06% on ImageNet. The code is available at: https://github.com/Nyquixt/CoConv.

In this paper, we analyze the performance of Multiplayer Cloud Gaming (MCG) systems. To that end, we introduce a model and new MCG-Quality of Service (QoS) metric that captures the freshness of the players' updates and fairness in their gaming experience. We introduce an efficient measurement-based Joint Multiplayer Rate Adaptation (JMRA) algorithm that optimizes the MCG-QoS by overcoming large (possibly varying) network transport delays by increasing the associated players' update rates. The resulting MCG-QoS is shown to be Schur-concave in the network delays, leading to natural characterizations and performance comparisons associated with the players' spatial geometry and network congestion. In particular, joint rate adaptation enables service providers to combat variability in network delays and players' geographic spread to achieve high service coverage. This, in turn, allows us to explore the spatial density and capacity of compute resources that need to be provisioned. Finally, we leverage tools from majorization theory, to show how service placement decisions can be made to improve the robustness of the MCG-QoS to stochastic network delays.

To reduce the amount of transmitted data, feature map-based fusion is recently proposed as a practical solution to cooperative 3-D object detection by autonomous vehicles (AVs). The precision of object detection, however, may require significant improvement, especially for objects that are far away or occluded. To address this critical issue for the safety of AVs and human beings, we propose a cooperative spatial feature fusion (CoFF) method for AVs to effectively fuse feature maps for achieving a higher 3-D object detection performance. Especially, CoFF differentiates weights among feature maps for a more guided fusion, based on how much new semantic information is provided by the received feature maps. It also enhances the inconspicuous features corresponding to far/occluded objects to improve their detection precision. The experimental results show that CoFF achieves a significant improvement in terms of both detection precision and effective detection range for AVs, compared to previous feature fusion solutions.

Motivated by the potentially high downlink traffic demands of commuters in future autonomous vehicles, we study a network architecture where vehicles use Vehicle-to-Vehicle (V2V) links to form relay network clusters, which in turn use Vehicle-to-Infrastructure (V2I) links to connect to one or more Road Side Units (RSUs). Such cluster-based multihoming offers improved performance, e.g., in coverage and per user shared rate, but depends on the penetration of V2V+V2I capable vehicles and possible blockage, by legacy vehicles, of line of sight based V2V links, such as those based on millimeter-wave and visible light technologies. This paper provides a performance analysis of a typical vehicle's connectivity and throughput on a highway in the free-flow regime, exploring its dependence on vehicle density, sensitivity to blockages, number of lanes and heterogeneity across lanes. The results, backed up by simulations of realistic vehicular traffic, show that even with moderate vehicle densities and penetration of V2V+V2I capable vehicles, such architectures can achieve substantial improvements in connectivity and reduction in per-user rate variability as compared to V2I based networks. The typical vehicle's performance is also shown to improve considerably in the multilane highway setting as compared to a single lane road. This paper also sheds light on how the network performance is affected when vehicles can control their relative positions, by characterizing the connectivity-throughput tradeoff faced by the clusters of vehicles.

The automotive industry is undergoing disruptive changes, e.g., ride sharing and self-driving cars which, in addition to leveraging wireless connectivity, may lead to dramatic changes in the volume of infotainment and work related data consumption of vehicle bound passengers. This paper studies the potential gains of leveraging clusters of V2V interconnected vehicles to enable: (1) improved opportunistic access to the cellular infrastructure; and (2), balancing traffic loads across cells through cluster multihoming. A stochastic geometric model and associated analysis are used to obtain a preliminary understanding of possible gains of cluster-based opportunistic relaying and its sensitivity to the system parameters, e.g., base station density, vehicular cluster size and density etc. An optimal network utility maximization formulation is then developed to serve as a baseline to evaluate a simple distributed cluster management algorithm which for the scenarios considered proves to be near-optimal. Overall the results suggest that 3-10x throughput gains are possible along with significant improvements in user rate fairness depending on the system parameters.

In this paper, we consider the problem of dynamic space-time resource allocation for optimizing the movements of connected autonomous vehicles (CAVs) through intersections without traffic signals. We design a three-dimensional (3D) space-time resource model for maintaining the intersection resource information in both the two-dimensional (2D) space domain and the time domain. In the 3D resource model, the trajectory of a CAV through an intersection is assigned a specific parallelepiped resource that spans both 2D space and time domains. Moreover, the dynamic space-time resource allocation problem is simplified to a classic 3D container-packing problem. We propose a dynamic heuristic algorithm, Best Parallelepiped Fit (BPF), to maintain smooth traffic flow and maximize spacetime resource usage by adjusting the speed and entry time of each approaching CAV through intersections. We evaluate the performance of the proposed algorithm under different traffic loads, and simulation results indicate that our algorithm can greatly reduce the average travel delay of CAVs.

In this paper, we study the survivable routing, modulation, and spectrum assignment (S-RMSA) problem for bulk data- flow transfer in elastic optical networks. We design a two- dimensional resource model for specifying resource state information that spans both the spectrum and time dimensions, and we develop a dynamic heuristic RMSA algorithm, Max Resource Utilization (MRU), to efficiently transfer data-flows while meeting deadline constraints and surviving single link failures. Furthermore, we provide a protection resource optimization strategy to progressively release the scheduled protection resources so as to increase spectrum utilization. We examine the performance of our proposed algorithm, and numerical results show that our proposed heuristic algorithm can reduce blocking probability and obtain high spectrum utilization while minimizing fragmentation.

We propose a multi-path fragmentation-aware routing, modulation and spectrum assignment algorithm (RMSA) for advance reservation (AR) and immediate reservation (IR) requests in elastic optical networks. To decrease fragmentation, we propose splitting requests into different parts and transferring each of these parts along a single-path or multi-paths utilizing sliceable bandwidth variable transponders. We first introduce a model to solve the problem and propose a two-dimensional fragmentation occurrence measurement in spectrum and time domains. Then we propose a multi-path fragmentation-aware RMSA algorithm (MPFA). Simulation results show that MPFA can achieve better performance than existing algorithms in terms of blocking probability and spectrum utilization.

In this paper, we consider the problem of routing, modulation, and spectrum assignment (RMSA) for data- flow transfers in elastic optical networks. We design a two-dimensional resource model, in which each data transfer with known data size can be assigned a rectangular block of resources that spans both the spectrum and time dimensions. Furthermore, the dynamic spectral resource allocation problem in elastic optical networks is simplified to the two-dimensional rectangle packing problem. We design a three-tuple for each rectangle placement, and develop a dynamic heuristic algorithm, Best Rectangle Fit (BRF), to efficiently schedule requests while minimizing fragmentation in both spectrum and time domains. We simulate the proposed algorithm, and the results show that the proposed RMSA algorithm (BRF) can greatly decrease blocking probability and increase spectrum utilization.

In this paper, we consider the problem of routing, modulation, and spectrum assignment (RMSA) for immediate and advance reservation requests in elastic optical networks. We design a two-dimensional resource model for maintaining resource state information in both spectrum and time domains. For the purpose of minimizing the blocking probability and increasing spectrum utilization, we develop a two phase RMSA algorithm that attempts to decrease spectrum resource fragmentation by scheduling the requests in a manner that takes into account the holding time of each request. We design a simulation to evaluate the performance of the proposed algorithm, and the experiment results show that our proposed two phase RMSA algorithm can greatly reduce blocking probability and obtain high spectrum utilization.

In this paper, we consider the problem of routing, modulation and spectrum assignment (RMSA) with holding time awareness in elastic optical networks. We consider a two dimensional schedule that maintains link resource usage in both spectrum and time domains, and we develop an RMSA algorithm that attempts to reduce resource fragmentation by scheduling requests in a manner that takes into account the holding time of each request. We examine the performance of the proposed algorithm through simulation experiments, and our results indicates that the proposed RMSA algorithm obtains high spectrum utilization and lower blocking probabilities compared to existing RMSA algorithms that do not make use of holding time information.