Ensemble models in E-commerce combine predictions from multiple sub-models for ranking and revenue improvement. Industrial ensemble models are typically deep neural networks, following the supervised learning paradigm to infer conversion rate given inputs from sub-models. However, this process has the following two problems. Firstly, the point-wise scoring approach disregards the relationships between items and leads to homogeneous displayed results, while diversified display benefits user experience and revenue. Secondly, the learning paradigm focuses on the ranking metrics and does not directly optimize the revenue. In our work, we propose a new Learning-To-Ensemble (LTE) framework RA-EGO, which replaces the ensemble model with a contextual Rank Aggregator (RA) and explores the best weights of sub-models by the Evaluator-Generator Optimization (EGO). To achieve the best online performance, we propose a new rank aggregation algorithm TournamentGreedy as a refinement of classic rank aggregators, which also produces the best average weighted Kendall Tau Distance (KTD) amongst all the considered algorithms with quadratic time complexity. Under the assumption that the best output list should be Pareto Optimal on the KTD metric for sub-models, we show that our RA algorithm has higher efficiency and coverage in exploring the optimal weights. Combined with the idea of Bayesian Optimization and gradient descent, we solve the online contextual Black-Box Optimization task that finds the optimal weights for sub-models given a chosen RA model. RA-EGO has been deployed in our online system and has improved the revenue significantly.

Recently, mobile ad-hoc networks (MANET) have been widely used in several scenarios. Due to its generally high demands on clock synchronization accuracy, the conventional synchronization algorithms cannot be applied in many high-speed MANET applications. Hence, in this paper, a clock synchronization algorithm based on motion information such as the speed of nodes is proposed to eliminate the error of round-trip-time correction. Meanwhile, a simplified version of our algorithm is put forward to cope with some resource-constrained scenes. Our algorithm can perform well in most situations and effectively improve the clock synchronization accuracy with reasonable communication overhead, especially in high-speed scenes. Simulation results confirm the superior accuracy performance achieved by our algorithm.

Channel estimation for underwater acoustic (UWA) channels in orthogonal frequency division multiplexing (OFDM) systems is one of the most difficult problems in UWA communications. Because UWA channels are both frequency-selective and time-selective, the total number of channel coefficients to be estimated will significantly increase. Besides, the fast time-varying characteristics and strong Doppler effects will deteriorate the orthogonality between the OFDM subcarriers and cause inter-carrier interference (ICI). To handle these problems, in this paper, we firstly transform the discrete doubly selective UWA channel model into an expansion of complex exponential basis. Based on the basis expansion model (BEM) of UWA channels, we then equivalently transform the channel estimation problem into the recovery of sparse BEM coefficients. Finally, a modified sparse signal recovery scheme based on block sparse Bayesian learning is proposed to estimate the doubly selective UWA channel state information (CSI). Simulation results confirm its performance merits.

In this paper, a spectral efficient channel estimation scheme is proposed for spatial-correlated sparse massive multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems with superimposed time-frequency training sequences (TSs) based on basis expansion model (BEM). This scheme first relies on the identical time-domain TS to acquire the partial channel common support, based on which the frequency-domain TSs inserted into the OFDM block are then utilized for accurate channel state information (CSI) acquisition. Besides, the superimposed TSs are adopted to reduce the TS overhead. Meanwhile, by exploiting the spatial correlations of massive MIMO channels, BEM can be utilized to reduce the number of the frequency-domain TSs and improve the spectral efficiency drastically from 22% to 95%. Simulation results show that the proposed scheme with 200 frequency-domain TSs can achieve almost the same performance as the conventional approach with 3200 frequency-domain TSs in massive MIMO systems with 256 transmit antennas.

In this study, a spectral efficient channel recovery scheme is proposed for spatial‐temporal correlated sparse massive multiple‐input multiple‐output (MIMO) orthogonal frequency division multiplexing systems. By exploiting the spatial‐temporal correlations of massive MIMO channels, superimposed time‐frequency training sequences (TSs) are jointly utilised for accurate channel state information (CSI) acquisition. Moreover, a novel time‐frequency TS pattern design scheme is proposed to improve the recovery performance based on structured compressive sensing theory. Furthermore, the discrete prolate spheroidal basis expansion model of massive MIMO channels in the space‐domain is utilised to reduce the frequency‐domain pilots overhead and improve the spectral efficiency. Simulation results show that the proposed scheme could achieve accurate CSI with great improvement in spectral efficiency and computational complexity.

Channel estimation for MIMO orthogonal frequency division multiplexing (MIMO-OFDM) systems has been improved observably based on compressive sensing (CS) techniques. Recently, some block-structured CS methods have been utilized in MIMO-OFDM systems. However, the block-coherence which characterizes the performance of a block-structured measurement matrix is still not optimized. In this paper, we proposed a two-step scheme to optimize the pilot locations and power of different transmitting antennas, in order to minimize the block-coherence of the aggregate system matrix. The selection of pilot locations and power can be obtained by two genetic algorithms. Then we introduce a block version of the orthogonal matching pursuit (OMP) algorithm, termed BOMP, to obtain the channel estimation consistent with the block-sparse model. Simulation results demonstrate that the proposed pilot allocation scheme with superimposed pilots design and block-structured CS method performs very well in MIMO-OFDM systems.