Split Federated Learning (SFL) enables scalable training on edge devices by combining the parallelism of Federated Learning (FL) with the computational offloading of Split Learning (SL). Despite its great success, SFL suffers significantly from the well-known straggler issue in distributed learning systems. This problem is exacerbated by the dependency between Split Server and clients: the Split Server side model update relies on receiving activations from clients. Such synchronization requirement introduces significant time latency, making straggler a critical bottleneck to the scalability and efficiency of the system. To mitigate this problem, we propose *MU-SplitFed*, a straggler-resilient SFL algorithm that decouples training progress from straggler delays via a simple yet effective unbalanced update mechanism. By enabling the server to perform $\tau$ local updates per client round, *MU-SplitFed* achieves convergence rate $\mathcal{O}(\sqrt{d/(\tau T)})$, showing a linear reduction in communication round by a factor of $\tau$. Experiments demonstrate that *MU-SplitFed* consistently outperforms baseline methods with the presence of stragglers and effectively mitigates their impact through adaptive tuning of $\tau$.

To address issues arising from the independence of the radio access network, the IP bearer network, and the data center network, a Cybertwin-based cloud native network (CCNN) is proposed. However, due to diverse application requirements and user demands, providing individualized transmission services that meet these requirements within CCNN remains a major challenge. This paper proposes a Cybertwin-enabled multipath transmission scheme (CMTS) to provide personalized quality-of-service (QoS)-guaranteed transmission services by dynamically creating network paths tailored to users’ demands. Specifically, by introducing Cybertwin, CMTS allows separate and dynamic scheduling of service and access network resources from different providers, decouples users from network providers, and enables more flexible use of heterogeneous network resources. Moreover, CMTS can make resource scheduling decisions on demand by leveraging the user’s personal information held by Cybertwin. To fully exploit the potential of multiple paths, we formulate network resource planning as an optimization problem aimed at minimizing link differences and propose a heuristic method for determining optimal policies. Finally, we present the first implementation of Cybertwin in CCNN, validate CMTS in both emulated and semi-physical environments, and conduct thorough evaluations with benchmarks. The results show that CMTS can deliver a personalized, QoS-guaranteed transmission service while achieving efficient resource utilization.

Haiquan Zhao, Lingyu Li, Shisong Chen, Shuqi Kong, Jiaan Wang, Kexin Huang, Tianle Gu, Yixu Wang, Jian Wang, Liang Dandan, Zhixu Li, Yan Teng, Yanghua Xiao, Yingchun Wang. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Powered by remarkable advancements in Large Language Models (LLMs), Multimodal Large Language Models (MLLMs) demonstrate impressive capabilities in manifold tasks.However, the practical application scenarios of MLLMs are intricate, exposing them to potential malicious instructions and thereby posing safety risks.While current benchmarks do incorporate certain safety considerations, they often lack comprehensive coverage and fail to exhibit the necessary rigor and robustness.For instance, the common practice of employing GPT-4V as both the evaluator and a model to be evaluated lacks credibility, as it tends to exhibit a bias toward its own responses.In this paper, we present MLLMGuard, a multi-dimensional safety evaluation suite for MLLMs, including a bilingual image-text evaluation dataset, inference utilities, and a lightweight evaluator.MLLMGuard's assessment comprehensively covers two languages (English and Chinese) and five important safety dimensions (Privacy, Bias, Toxicity, Truthfulness, and Legality), each with corresponding rich subtasks.Focusing on these dimensions, our evaluation dataset is primarily sourced from platforms such as social media, and it integrates text-based and image-based red teaming techniques with meticulous annotation by human experts.This can prevent inaccurate evaluation caused by data leakage when using open-source datasets and ensures the quality and challenging nature of our benchmark.Additionally, a fully automated lightweight evaluator termed GuardRank is developed, which achieves significantly higher evaluation accuracy than GPT-4.Our evaluation results across 13 advanced models indicate that MLLMs still have a substantial journey ahead before they can be considered safe and responsible.

Space-air-ground integrated networks (SAGIN) have been envisioned as the promising and key network architecture for the 6G vehicular networks to provide seamless coverage for the connected vehicles. To access the most appropriate network quickly, this paper proposed a knowledge-driven network access approach, where the communication knowledge is explicitly integrated into neural networks, to deal with multiple tasks in SAGIN-based vehicular networks. Specifically, the formulated long-term network access problem is handled by asynchronous advantage actor-critic algorithm (A3C) in reinforcement learning. During this process, the space-time correlation knowledge is introduced to effectively reduce the action space in channel selection and the reward shaping exploiting the problem-specific communication and mathematical knowledge is adopted to solve the sparse reward problem in reinforcement learning. In addition, by modifying the sub-net learning rate of the A3C algorithm with experimental experience, this paper speeds up the network convergence speed by 1.5%. Numerical results also show that integrating knowledge into traditional deep reinforcement learning algorithm can improve the reward by 4%.

MOTIVATION: The metabolome and microbiome disorders are highly associated with human health, and there are great demands for dual-omics interaction analysis. Here, we designed and developed an integrative platform, 3MCor, for metabolome and microbiome correlation analysis under the instruction of phenotype and with the consideration of confounders. RESULTS: Many traditional and novel correlation analysis methods were integrated for intra- and inter-correlation analysis. Three inter-correlation pipelines are provided for global, hierarchical and pairwise analysis. The incorporated network analysis function is conducive to rapid identification of network clusters and key nodes from a complicated correlation network. Complete numerical results (csv files) and rich figures (pdf files) will be generated in minutes. To our knowledge, 3MCor is the first platform developed specifically for the correlation analysis of metabolome and microbiome. Its functions were compared with corresponding modules of existing omics data analysis platforms. A real-world dataset was used to demonstrate its simple and flexible operation, comprehensive outputs and distinctive contribution to dual-omics studies. AVAILABILITYAND IMPLEMENTATION: 3MCor is available at http://3mcor.cn and the backend R script is available at https://github.com/chentianlu/3MCorServer. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Metabolomics data analyses rely on the use of bioinformatics tools. Many integrated multi-functional tools have been developed for untargeted metabolomics data processing and have been widely used. More alternative platforms are expected for both basic and advanced users. RESULTS: Integrated mass spectrometry-based untargeted metabolomics data mining (IP4M) software was designed and developed. The IP4M, has 62 functions categorized into 8 modules, covering all the steps of metabolomics data mining, including raw data preprocessing (alignment, peak de-convolution, peak picking, and isotope filtering), peak annotation, peak table preprocessing, basic statistical description, classification and biomarker detection, correlation analysis, cluster and sub-cluster analysis, regression analysis, ROC analysis, pathway and enrichment analysis, and sample size and power analysis. Additionally, a KEGG-derived metabolic reaction database was embedded and a series of ratio variables (product/substrate) can be generated with enlarged information on enzyme activity. A new method, GRaMM, for correlation analysis between metabolome and microbiome data was also provided. IP4M provides both a number of parameters for customized and refined analysis (for expert users), as well as 4 simplified workflows with few key parameters (for beginners who are unfamiliar with computational metabolomics). The performance of IP4M was evaluated and compared with existing computational platforms using 2 data sets derived from standards mixture and 2 data sets derived from serum samples, from GC-MS and LC-MS respectively. CONCLUSION: IP4M is powerful, modularized, customizable and easy-to-use. It is a good choice for metabolomics data processing and analysis. Free versions for Windows, MAC OS, and Linux systems are provided.

An attractive hybrid method of mitigating the effects of the residual self-interference imposed by the signal propagation and analog/digital circuit non-idealities for Full-Duplex (FD) point-to-point multiple-input multiple-output (MIMO) systems is proposed. Furthermore, the effect of channel estimation errors on system performance is considered. The simulation results demonstrate that our proposed cancellation technique for FD systems achieve a significant gain over traditional Half-Duplex (HD) systems. For example, the sum rate of our proposed scheme of FD MIMO system is approximately 1.8 times higher than that of the HD transmission.

A novel adaptive turbo-coded soft-decision aided differential detection (ATSDD) scheme is proposed for cooperative uplink wireless and Radio-over-fiber (ROF) transmission in a Fractional Frequency Reuse (FFR) based multicell, multiuser system. More specifically, the ATSDD scheme is employed by the Mobile Station (MS) for reliably conveying the source bits to a pair of Remote Antennas (RAs) by appropriately adjusting the modulation mode according to the near-instantaneous wireless and ROF channel conditions. The ATSDD switching thresholds are specifically adjusted for ensuring that the Bit Error Ratio (BER) remains below 10^(-5). We also investigated the effect of phase-rotations, which is routinely inflicted by practical imperfect ROF links. We demonstrate that our ATSDD scheme increases the overall throughput.

Multilevel Differential Amplitude and Phase-Shift Keying (DAPSK) schemes do not require any channel estimation, which results in low complexity. In this treatise we derive the soft-output probability formulas required for a soft-decision based demodulation of high-order DAPSK, in order to facilitate iterative detection by exchanging extrinsic information with an outer Turbo Code (TC). Furthermore, when the TC block size is increased, the system operates closer to the channel capacity. Compared to the identical-throughput TC assisted 64-ary Differential Phase-Shift Keying (64-DPSK) scheme, the 4-ring based TC assisted 64-ary DAPSK arrangement has a power-efficiency improvement of 2.3 dB at a bit error rate (BER) of 10-5. Furthermore, when the TC block size is increased, the system operates closer to the channel capacity. More specifically, when using a TC block length of 400 modulated symbols, the 64 DAPSK (4, 16) scheme is 7.56 dB away from its capacity curve, while it had a reduced gap as low as 2.25 dB, when using a longer TC block length of 40 000 modulated symbols. Finally, as a novel application example, the soft-decision M-DAPSK scheme was incorporated into an Amplify-and-Forward (AF) based cooperative communication system, which attains another 4.5 dB SNR improvement for a TC block length of 40 000 modulated symbols.

In this paper, we propose to reduce the complexity of both the Approx-Log-MAP algorithm as well as of the Max-Log-MAP algorithm conceived for generalized PSK/QAM detection, where only a reduced-size subset of the PSK/QAM constellation points is taken into account for producing a single soft-bit output. Although the detectors of Gray-labelled low-order PSK/QAM schemes generally produce near-horizontal EXIT curves, our proposed detectors exploit the a priori LLRs gleaned from a channel decoder in order to retain the optimum detection capability for all PSK/QAM constellations. Furthermore, we demonstrate in this paper that the widely applied MIMO schemes including V-BLAST and STBC, which invoke the proposed soft PSK/QAM detectors may also benefit from our reduced-complexity design. Our simulation results confirm that a near-capacity performance may be achieved by the proposed detectors at a substantially reduced detection complexity.

A multilayered surface coil array for magnetic resonance imaging with an improved signal-to-noise ratio (SNR) performance is introduced and investigated by a simulation study. By using an effective decoupling method, the strong mutual coupling effect between the coil layers can be accurately removed, leading to a coherent combination of the signals of the individual coils. This results in a much stronger received signal power which increases with the number of coil layers in the array. This, together with a smaller rate of increase of noise power with the number of coil layers, leads to a net increase in the SNR of array output with the number of coil layers in the array. Rigorous numerical simulation examples have been carried out to confirm and verify the performance of the new array.

In this paper, we propose to reduce the complexity of both the Approx-Log-MAP algorithm as well as of the Max-Log-MAP algorithm, which were designed for soft-decision-aided Space-Time Block Code (STBC) detectors. First of all, we review the STBC design, which enables regular L-PSK/QAM detectors to be invoked in order to detect STBCs on a symbol-by-symbol basis. Secondly, we propose to operate the L-PSK/QAM aided STBC detection on a bit-by-bit basis, so that the complexity may be reduced from the order of O(L) to O(BPS = log2L). Our simulation results demonstrate that a near-capacity performance may be achieved by the proposed detectors at a substantially reduced detection complexity. For example, a factor six complexity reduction was achieved by the proposed algorithms, when they were invoked for detecting Alamouti's Square 16QAM aided G2 scheme.

In this paper, we propose to reduce the complexity of both the Approx-Log-MAP algorithm as well as of the Max-Log-MAP algorithm, which were designed for soft-decision aided PSK detectors. First of all, we extend the shown a posteriori PSK symbol probability formula and streamline it by eliminating its unnecessary calculations in the context of the Approx-Log-MAP algorithm. Secondly, we reduce the complexity of the Max-Log-MAP algorithm, where the maximum a posteriori symbol probability may be obtained without evaluating and comparing all the candidate symbol probabilities. Furthermore, we apply our new soft detection arrangement to a variety of coded systems. Our simulation results demonstrate that a significant detection complexity reduction was achieved by our design without any performance loss. For example, a factor two complexity reduction was achieved by the proposed Max-Log-MAP algorithm, when it was invoked for detecting QPSK symbols, which is expected to be significantly higher, when invoked for 16QAM.

A new method is introduced to increase the signal-to-noise ratio (SNR) in low-field magnetic resonance imaging (MRI) systems by using a vertically stacked phased coil array. It is shown theoretically that the SNR is increased with the square root of the number of coils in the array if the array signals are properly combined to remove the mutual coupling effect. Based on this, a number of vertically stacked phased coil arrays have been designed and characterized by a numerical simulation method. The performance of these arrays confirms the significant increase of SNR by increasing the number of coils in the arrays. This provides a simple and efficient method to improve the SNR for low-field MRI systems.

Star Quadrature Amplitude Modulation (Star-QAM) is a non-coherent detection aided scheme ideal for low-complexity wireless transceivers. In this contribution, we first derive soft-decision aided StQAM symbol-to-bit demapping combined with iterative detection aided Turbo Coding (TC). Then we combine this physical layer design with network coding (NC) and quantify the attainable coding gain. It is shown that the proposed 16-level Star-QAM (16-StQAM) based TC scheme is capable of achieving a coding gain of about 1.5~dB over the TC aided 16-level Differential Phase-Shift Keying (16DPSK) benchmark scheme at a BER of 10-5. The 16-StQAM based TC assisted scheme is capable of offering another 1.2 dB coding gain, when it is employed in a 'butterfly' topology network arrangement. When the source and relay nodes are allowed to transmit at different power levels, the two-hop 16-StQAM based TC aided NC scheme outperforms the single-hop 16-StQAM based TC scheme by approximately 2.2 dB.

Low-complexity non-coherently detected Differential Amplitude and Phase-Shift Keying (DAPSK) schemes constitute an ideal candidate for wireless communications. In this paper, we derive the soft- output probability formulas required for the soft- decision based demodulation of DAPSK, which are then invoked for Turbo Coded (TC) transmissions. Furthermore, the achievable throughput characteristics of the family of M-ary DAPSK schemes are provided. It is shown that the proposed 4-ring based TC assisted 64-ary DAPSK scheme achieves a coding gain of about 4.2 dBs in comparison to the identical-throughput TC assisted 64-ary Differential Phase-Shift Keying (64-DPSK) scheme at a bit error ratio of 10-5.

Differentially detected noncoherent Star Quadrature Amplitude Modulation (Star-QAM) is ideal for low-complexity wireless communications, since it dispenses with high-complexity channel estimation. We conceive soft-decision based demodulation for 16-level Star-QAM (16-StQAM), which is then invoked for iterative detection aided Iteratively-Detected Bit-Interleaved Coded Modulation (BICM-ID). It is shown that the proposed 16-StQAM based BICM-ID scheme achieves a coding gain of approximately 14 dBs in comparison to the 16-level identical-throughput Differential Phase-Shift Keying (16DPSK) assisted BICM scheme at a bit error ratio of 10-6.

An attractive hybrid method of mitigating the effects of error propagation that may be imposed by the relay node (RN) on the destination node (DN) is proposed. We selected the most appropriate relay location for achieving a specific target Bit Error Ratio (BER) at the relay and signalled the RN-BER to the DN. The knowledge of this BER was then exploited by the decoder at the destination. Our simulation results show that when the BER at the RN is low, we do not have to activate the RN-BER aided decoder at the DN. However, when the RN-BER is high, significant system performance improvements may be achieved by activating the proposed RN-BER based decoding technique at the DN. For example, a power-reduction of up to about 19dB was recorded at a DN BER of 10-4.

Abstract-In this contribution, we propose an adaptive Turbo Trellis Coded Modulation (TTCM) aided Distributed Space-Time Trellis Coding (STTC) scheme for cooperative communication over quasi-static Rayleigh fading channels. An adaptive TTCM (ATTCM) scheme is employed by the source node during the first transmission period for reliably conveying the source bits to N number of relay nodes by appropriately adjusting the code-rate and modulation mode according to the near-instantaneous channel condition. The TTCM switching thresholds are chosen to ensure that the Bit Error Ratio (BER) at each relay node becomes lower than 10-6in order to minimise the potential error propagation imposed by the relay nodes. During the second transmission period, an N-antenna assisted Distributed STTC (DSTTC) scheme is created with the aid of the aforementioned N single-antenna relay nodes. More specifically, the N relay nodes are utilised to form STTC codewords based on the re-encoded TTCM symbols of the relays for transmission to the destination node. At the destination node, iterative extrinsic information exchange is performed between the STTC and TTCM decoders for recovering the original source bits. It is shown that the proposed ATTCM-DSTTC scheme requires 12 dBs less transmission power in comparison to a standard TTCM scheme when aiming for a frame error ratio of 10-3.