Differentiable architecture search (DARTS) suffers from over-parameterized supernetworks that make architecture search difficult to perform in complex detection scenarios. In this paper, a shareable attention-mask for object detection architecture search (SAM-Det) is proposed, which is designed to address the high computational cost and poor performance caused by search over-parameterization. To reduce the search cost, a series of convolutional candidate operations is merged by using a shareable attention mask to represent the operations as a single convolution with shared weights. Moreover, to reduce the interference of redundant features on the search, the features are shuffled in groups to learn the attentional importance, thus selecting the important channels to be transferred into the search space. The experimental results show that SAM-Det is able to obtain network architectures with better detection performance and lower search cost in different detection scenarios compared to other Search methods.

Photovoltaic (PV) power forecasting is important for promoting the integration of renewable energy sources. However, neural network-based methods, particularly deep learning for PV power forecasting, face challenges with computational and memory requirements when dealing with industry-scale datasets. To address this, we introduce Rec-PF, a robust computational framework employing the tensor-train (TT) technique. This framework aims to streamline the training process of massive deep learning recommendation models (DLRMs) on constrained resources. Rec-PF employs a high-performance compressed embedding table, enhancing TT decomposition using key computing primitives. It serves as a drop-in replacement for the PyTorch API. Additionally, Rec-PF utilizes an index reordering technique to leverage local and global information from training inputs, thereby enhancing performance. Furthermore, Rec-PF adopts a pipeline training model, eliminating the need for communication between training workers and host memory. We are pioneers in applying DLRM to PV power prediction to reduce training time without compromising accuracy. Our approach demonstrates a twofold improvement in training time compared to methods that do not incorporate our approach. To better demonstrate the enhanced performance of the algorithm, we specifically compare its efficiency with other frameworks using datasets commonly employed in recommender systems. Comprehensive experiments indicate that Rec-PF is capable of processing the largest publicly accessible DLRM and PV datasets on a single GPU, offering a threefold acceleration compared to state-of-the-art DLRM and PV frameworks.

This paper develops an analytical framework for exploring the benefits of applying cluster-based multi-antenna non-orthogonal multiple access (NOMA) in dense wireless networks. Using the tools of stochastic geometry, a new explicit expression and a tight approximation for per-cluster average data rates are derived in terms of relevant system parameters. Moreover, simulation results are provided to validate the accuracy of our analytical results and draw some essential system design insights. Based on the tractable expressions, we further consider the analysis and optimization of area spectral efficiency (ASE). It is analytically demonstrated that: 1) In terms of the per-cluster average data rate, there exists a wide range for the power allocation coefficient within a cluster where NOMA outperforms orthogonal multiple access (OMA); 2) The per-cluster average data rate of NOMA benefits more from the additional deployment of antennas at base stations (BSs) than that of OMA when the accuracy of channel state information at transmitter (CSIT) is fixed; 3) Regarding to ASE, there exists an optimal combination of the number of clusters and power allocation coefficient to maximize ASE. The performance gain of NOMA relative to OMA becomes marginal when the requirement of user fairness is stringent; 4) As for a special case that the power allocation coefficient is fixed, the analytical results indicate that ASE scales linearly with the number of antennas when the number of clusters is set optimally.

From the perspective of statistical performance, this paper presents a framework for the per-user throughput analysis in random cache based multi-antenna heterogeneous networks (HetNets) with user-centric inter-cell interference nulling (IN). Using tools from stochastic geometry, an explicit expression for the per-user throughput is derived. Based on the analytical results, the optimal cache probabilities for maximizing the per-user throughput are analyzed. Theoretical analysis and numerical results reveal that the optimal random cache under interference nulling fully harvests the file diversity gain (FDG) and achieves a promising per-user throughput.

In this letter, we propose a new channel estimation scheme for downlink channels in massive multiple-input multiple-output systems, where orthogonal frequency-division multiplexing is adopted. To estimate the downlink channels in the multi-subcarrier scenario, the common sparsity and cluster structure is exploited, which is unknown to the user. The common sparsity property is described and a local beta process is assumed on each of the common local clusters in a new constructed Bayesian framework. Then, we propose a common structure based multi-subcarrier Bayesian compressive sensing approach for the downlink channel estimation. Simulation results verify the effectiveness of the proposed algorithm.