This paper aims to address the subject-dependent challenge of the facial emotion recognition (FER) task. To accomplish this, we propose a novel face recognition based attention FER (FRA-FER) framework which propagates subtle face recognition (FR) features through the FER network. Particularly, first a spatial attention map from the feature maps of an FR convolutional neural network (CNN) is created and then it is fused into the FER-CNN. By doing this FR feature propagation, the FER network is personalized as it takes the advantage of the FR features learned from large-scale face recognition datasets. Experiments on the two challenging datasets AffectNet and AFEW demonstrate the superiority of our proposed FRA-FER network to the state-of-the-art work.

Locally repairable codes (LRCs) have been recently proposed and used in real-world distributed storage systems (DSSs) such as Microsoft Azure Storage and Facebook HDFS-RAID (Hadoop Distributed File System-Redundant Array of Independent Disks). Since information in DSSs is changed frequently, reducing update complexity (UC) of LRCs is of great interest. In this paper, we propose code design algorithms that can reduce UC of existing LRCs without sacrificing their important code parameters such as minimum distance, code rate, or locality. We establish bounds on UC, and use them to show that our algorithms can achieve optimal or near optimal UC for a large class of LRCs.

A linear block code with dimension k, length n, and minimum distance d is called a locally repairable code (LRC) with locality r, if it can retrieve any coded symbol by at most r other coded symbols. LRCs have been recently proposed and used in practice in distributed storage systems, such as Windows Azure Storage and Facebook HDFS-RAID. Theoretical bounds on the maximum locality of LRCs (r) have been established. The average locality of an LRC (r) directly affects the costly repair bandwidth, disk I/O, and the number of nodes involved in the repair process of a missing data block. There is a gap in the literature studying r. In this paper, we establish a lower bound on r of arbitrary (n, k, d) LRCs. Furthermore, we obtain a tight lower bound on r for a practical case where the code rate (R = (k/n)) is greater than (1 - (1/√n))2. Finally, we design three classes of LRCs that achieve the obtained bounds on r. Comparing with the existing LRCs, our proposed codes improve the average locality without sacrificing such crucial parameters as the code rate or minimum distance.

Locally repairable codes (LRCs) have been proposed and used in practice as effective coding methods for distributed storage systems (DSSs). In a DSS, information block recovery is a critical task performed in the case of data node permanent failure or temporal unavailability. Temporal node unavailability accounts for 90% of all block recoveries triggered in DSS. Since parity blocks are not needed to be recovered during a temporal node unavailability, special attention should be given to reconstruction of information blocks when trying to minimize the average bandwidth needed for block recovery. Motivated by this, in this work, we study the average locality ofinformation blocks. We obtain a lower bound on the average locality of information blocks of LRCs and design LRCs that achieve the bound. In addition to obtaining the optimal average locality for the information blocks, our codes achieve the optimal maximum locality for all the information blocks as well as some parity blocks (in some cases all the parity blocks).

Due to their superior performance and moderate computational complexity, quasi‐orthogonal space–frequency block codes (QOSFBCs) and quasi‐orthogonal space–time–frequency block codes (QOSTFBCs) are among prime candidates for implementation in multiple‐input multiple‐output communications systems. Originally, these codes were introduced for two transmit antennas and it is proved that they are full‐diversity within channels with equal‐power profile and integer delays. In this study, first, the authors generalise the construction of QOSFBCs and QOSTFBCs for arbitrary number of transmit antennas. Then, they provide the theoretical proof that these codes are full‐diversity for any arbitrary number of transmit antennas and any arbitrary delay and power profiles. Furthermore, they prove that the coding advantage of these codes is decomposed into two distinct parts, one of which represents the effect of the channel and the other represents the effect of the precoder. By optimising these two parts, they propose a modified version of QOSFBCs and QOSTFBCs which outperforms the latest space–frequency block codes and space–time–frequency block codes in the literature according to the analytical results and support of simulation results. Finally, they propose a new suboptimum linear decoder for the QOSFBCs and the QOSTFBCs which could achieve almost the same performance as the maximum‐likelihood decoder for large number of subcarriers.

An (n, k) erasure code that can recover any coded symbol by at most r other coded symbols is called a locally repairable code (LRC) with locality r. LRCs have been recently implemented in distributed storage systems. Coding complexity reduction can be significantly decreased by using binary LRCs (BLRCs) as they eliminate costly multiplication calculation. In this paper, motivated by the recently erasure codes with d = 4 used in practice, we propose BLRCs when (r + 1) | n and d = 4. We prove that our proposed binary codes are optimal for r ∈ {1, 3}, meaning that neither their locality nor their minimum distance can be improved by non-binary codes. For r ≥ 4, our proposed binary codes offer near-optimal code rate, with a rate gap of O(log r/n) compared with optimal nonbinary codes. While keeping the bulk of code structure binary, we eliminate this rate gap by using fields with sizes as small as r + 2 for only two redundant symbols. These non-binary codes still eliminate the need for costly multiplications in many operations including a single failure repair (a dominant repair scenario). Using the construction of spanning BLRC with d = 4 as a backbone, we also construct LRCs with minimum distance d ≥ 6. Furthermore, we obtain a closed-form equation for the mean-time to data-loss of arbitrary erasure codes.

Motivated by the statistical correspondence between phases of the transmitted and received vectors, we present a fast decoding method for quasi-orthogonal space-time block codes (QOSTBCs) in this paper. The proposed decoder selects proper candidates from precomputed and sorted sets by focusing on the phase of a specific entry of the combined and decoupled vector. The ML metric of the most probable candidates is first evaluated, and then, the remaining candidates are assessed based on the similarity between the phases. The new algorithm can work with any type of constellation such as QAM and PSK and supports generalized block-diagonal QOSTBCs. Theoretical results backed by simulation tests confirm the solution's superior performance in terms of the decoder complexity compared with popular approaches found in the literature. This feature is achieved through far fewer computations operations compared with many of the state-of-the-art models reported in the literature.

In this study, the authors show that the coding advantage (CA) of quasi‐orthogonal space–frequency block codes (QOSFBCs) could be decomposed into two parts, namely intrinsic CA (ICA) and extrinsic CA (ECA). Then, by using the ICA, the authors demonstrate that the optimum precoder of the QOSFBCs could be analytically derived without any requirement of the exhaustive search. Then, by improving the ECA, the authors enhance performances of the QOSFBCs efficiently – up to 4 dB. Finally, the authors show that a very fast optimisation process for the QOSFBCs is viable over any arbitrary frequency‐selective channel.