CZChongsheng Zhang

Chongsheng Zhang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Chongsheng Zhang has a total of 83 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Real-Time 2-D-to-1-D Decomposition of Roesser Models via Sylvester Equation Transformation With Interval-Constrained Balanced Model Reduction

    by Muhammad Imran, Hira Ambreen, Gaojuan Fan, Chongsheng Zhang

    The complexity of 2-D systems makes them challenging to analyze and manage. Basic algebraic operations are more intricate in 2-D frameworks than in their 1-D counterparts. Standard 2-D systems can typically only be decomposed into 1-D subsystems if they are represented in a separable denominator form. This research introduces a novel transformation that facilitates the decomposition of a 2-D standard system into a diagonalized 2-D representation; this further enables its partitioning into two 1-D subsystems, bypassing the constraints of minimal rank decomposition criteria. The proposed framework (PF) retains the geometrical symmetry of the resultant 1-D submodels while maintaining balance across the submodels, making them suitable for model reduction in large-scale systems. Furthermore, a novel framework for model reduction is developed to address limitations in existing limited-interval (LI) methods, such as instability within the reduced system. The numerical findings indicate that the proposed transformations adequately diagonalize the standard 2-D system and ensure a stable reduced 2-D model.


  • Multi-Modal Character Localization and Extraction for Chinese Text Recognition

    by Qilong Li, Chongsheng Zhang

    Scene text recognition (STR) methods have demonstrated their excellent capability in English text images. However, due to the complex inner structures of Chinese and the extensive character categories, it poses challenges for recognizing Chinese text in images. Recently, studies have shown that the methods designed for English text recognition encounter an accuracy bottleneck when recognizing Chinese text images. This raises the question: Is it appropriate to apply the model developed for English to the Chinese STR task? To explore this issue, we propose a novel method named LER, which explicitly decouples each character and independently recognizes characters while taking into account the complex inner structures of Chinese. LER consists of three modules: Localization, Extraction, and Recognition. Firstly, the localization module utilizes multimodal information to determine the character's position precisely. Then, the extraction module dissociates all characters in parallel. Finally, the recognition module considers the unique inner structures of Chinese to provide the text prediction results. Extensive experiments conducted on large-scale Chinese benchmarks indicate that our method significantly outperforms existing methods. Furthermore, extensive experiments conducted on six English benchmarks and the Union14M benchmark show impressive results in English text recognition by LER.

  • Distance-Gradient-Based Convex Optimization for Efficient Near-Optimal Coverage in WSNs

    by Gaojuan Fan, Feitao Li, Chongsheng Zhang, Hafiz Muhammad Sanaullah Badar, Christian Heumann, Ke Yuan

    Coverage optimization in wireless sensor networks (WSNs) is a fundamental yet NP-hard problem that directly affects monitoring quality and efficiency. Existing solutions mainly rely on meta-heuristic algorithms that use fitness-based evaluations, which often incur high computational overhead, slow convergence, and limited scalability, particularly in real-time or high-precision monitoring scenarios. In this article, we examine the relationship between effective coverage area and redundant distances in an analytical manner. We then propose reformulating WSN coverage optimization as a distance-gradient-based convex optimization problem, which can be subsequently solved using the first-order gradient descent (GD) algorithm or the second-order quasi-Newton algorithm. Extensive comparative experiments against five representative meta-heuristic methods, the virtual force algorithm (VFA) and a general convex optimization algorithm (CVX), demonstrate that our approach achieves near-optimal coverage while preserving network connectivity within milliseconds, highlighting its advantages over existing methods for WSNs coverage optimization.

  • Anomal-EFD: A self-supervised model for anomaly detection in dynamic IoT networks

    by Gaojuan Fan, Qingyi Huang, Junchao Ma, Badar Hafiz Muhammad Sanaullah, Chongsheng Zhang, Ke Yuan


  • ATTA-FL-Lite: Lightweight Byzantine-Robust Federated Learning for Resource-Constrained Medical IoT Devices

    by Hafiz Muhammad Sanaullah Badar, Nadeem Iqbal Kajla, Khalid Mahmood, Muhammad Waseem Khan, Gaojuan Fan, Chongsheng Zhang

    Federated learning (FL) enables privacy-preserving analytics at the medical IoT (IoMT) edge but is vulnerable to model poisoning and distribution shift. We present ATTA-FL-Lite, a lightweight aggregation rule that admits a client update only when three tests are jointly satisfied: (i) scale conformity via a median–absolute–deviation (MAD) z-score, (ii) directional alignment via cosine similarity to a coordinate-wise median reference, and (iii) non-degradation via a validation-loss z-score computed on a small, centrally held clean set. If no update passes, a coordinate-wise median fallback is used. We provide sub-Gaussian tail bounds for the loss test and an expected one-step descent bound for L-smooth objectives under benign mean-alignment and an accepted-set composition assumption; the analysis does not require a positive cosine threshold. Experiments on MNIST, Fashion-MNIST, and PathMNIST, under IID and non-IID partitions with up to 40% adversaries across four attack families, show that ATTA-FL-Lite maintains accuracy representatively ≈ 0.78–0.98 across Tiny/Small/Medium CNNs, reliably filters magnitude/noise attacks, and remains competitive against sign-flip. Server runtime scales approximately linearly with the number of participating clients at fixed model and validation sizes. These results indicate that ATTA-FL-Lite offers practical robustness for FL in resource-constrained IoMT deployments without cryptographic overhead or trusted root data beyond a small validation set.


  • Edge-Optimized Lightweight and Transformer Backbones for Real-Time Road Damage Detection in IIoT Systems

    by Hafiz Muhammad Sanaullah Badar, Israr Hussain, Ali Kashif Bashir, Nazik Alturki, Gaojuan Fan, Chongsheng Zhang

    Accurate and efficient road damage detection is critical for maintaining urban infrastructure and ensuring public safety in Intelligent Internet of Things (IIoT) systems. There remains a significant challenge to achieving a balance between detection accuracy and real-time inference on resource-constrained edge devices despite advances in deep learning. This article addresses this gap by enhancing the YOLOv9c object detection framework with two distinct backbone architectures: MobileNet V3-Small, which is a lightweight convolutional neural network optimized for edge deployment, and Swin Transformer, which is a hierarchical vision transformer that captures rich contextual features. We present a systematic, dual-backbone performance benchmark that quantifies the critical trade-off between computational efficiency and detection precision, which is essential for guiding IIoT deployment strategies. We conducted experiments on the street view road damage detection (SVRDD) dataset to evaluate detection accuracy, computational efficiency, and latency. The MobileNet backbone achieves the highest mean average precision (mAP@0.5) of 74.0% (a 1.5% gain over baseline) and recall of 68.8% (a 7.3% gain over baseline), demonstrating improved accuracy while maintaining a low inference time on a baseline GPU, indicating its suitability for deployment on IoT edge devices. Importantly, the MobileNet variant reduces the parameter count from 25.6 M to 2.54 M and the Giga floating-point operations per second (GFLOPs) from 102.3 to 0.49, making it more efficient for IIoT edge devices. Both backbones performed better than the YOLOv9c baseline model in terms of accuracy, thus providing scalable and practical solutions for real-time infrastructure monitoring. These findings contribute to the development of intelligent, efficient, and scalable object detection systems tailored for smart city and IIoT environments.


  • PoisonShield-FL-NIDS: A Robust Defense Against Poisoning Attacks in Federated Learning Intrusion Detection

    by Muhammad Ilyas, Nadeem Iqbal Kajla, Michael G. Madden, Gaojuan Fan, Chongsheng Zhang, Hafiz Muhammad Sanaullah Badar

    Federated learning (FL) has emerged as a privacy-preserving paradigm for collaborative intrusion detection in networked environments. However, it remains vulnerable to poisoning attacks (PAs) wherein malicious clients can corrupt the global model through deceptive updates. To address this, we propose PoisonShield-FL-network intrusion detection systems (NIDS), a robust FL-based intrusion detection system that integrates client-side anomaly filtering with trust-aware aggregation to defend against poisoned contributions. Experimental evaluation under varying levels of adversarial influence demonstrates that PoisonShield-FL-NIDS achieves superior performance across key metrics, attaining 93% accuracy, 91% precision, 94% recall, and an AUC of 0.96, while maintaining a low robustness index $ \text {RI}~\lt 0.05 $ even with 30% compromised clients. Compared to baseline FL models such as FL-CNN and FedACNN, our framework demonstrates faster convergence and higher resilience with only a marginal increase in communication overhead.


  • Open Access

    Complexity-Aware Deep Learning Framework for Intrusion Detection in Resource-Constrained Networks

    by Matthew Iwada Ekum, Idris Abiodun Aremu, Chongsheng Zhang, Pearse Oludare Adegbayi, Albert Ibikunle Idowu

    Artificial intelligence (AI) has emerged as a powerful tool for securing modern networks against increasinglysophisticated cyberattacks. However, many existing AI-basedintrusion detection systems (IDS) prioritise detection accuracywhile neglecting computational complexity, thereby limiting theirapplicability in resource-constrained environments such as Internet of Things (IoT) devices, edge computing nodes, and embeddedsystems. This paper presents a complexity-aware deep learningframework that explicitly integrates algorithmic complexity considerations into the design, feature selection, and learning stagesof IDS models. Rather than introducing a new deep architecture,the framework achieves a favourable accuracy–efficiency tradeoff by coupling gradient-based dynamic feature pruning witha lightweight neural network, supported by theoretical timeand space complexity analysis. Experiments on two benchmarkdatasets (NSL-KDD and CICIDS2017) demonstrate that theproposed method matches or surpasses the accuracy of stateof-the-art models while significantly reducing FLOPs, parameters, and inference cost. The study bridges the gap betweeninformation security, artificial intelligence, and computationalcomplexity theory, offering a simple, theoretically grounded,and reproducible pathway for deploying secure AI systems inresource-constrained environments.

  • Open Access

    Robust methylome analysis and tumour–normal classification in TCGA–COAD: a reproducible workflow

    by Matthew Iwada Ekum, Christian Heumann, Eno Emmanuella E. Akarawak, Khalid Olajide Adekoya, Segun Isaac Oke, Adeyinka Solomon Ogunsnaya, Chongsheng Zhang, Olalekan Joseph Akintande, Joseph Adekunle Akinyemi, Isaac Ajiboye, et al.

    Colorectal adenocarcinoma is caused in part by widespread epigenetic deregulation, yet the analysis of genome-wide DNA methylation of colorectal adenocarcinoma is complicated due to spatial correlation among CpG, multiscale patterns of differential methylation, and confounding cellular heterogeneity in bulk tissue. This study develops a simple yet effective framework that combines rigorous statistical modelling with modern machine learning-based prediction, on the Illumina 450K data from The Cancer Genome Atlas (TCGA) colorectal adenocarcinoma cohort. In our framework, differentially methylated regions (DMRs) were first detected using functional smoothing and permutation-based bump hunting, revealing both focal CpG island hypermethylation and broad hypomethylated domains spanning hundreds of kilobases. Next, we performed reference-based cell-type deconvolution and surrogate variable analysis (SVA) controlled immune/stromal admixture and hidden confounding effects, yielding well-calibrated single-site and region-level inference. Then, for tumour-status prediction, we empirically study the performance of classical Logistic Regression model against Random Forest, Gradient Boosting (XGBoost), and Feed-forward neural network; the results show that Logistic Regression achieves the lowest root-mean-square error and Brier score, reflecting its superior probability calibration. In general, our integrated framework provides biologically interpretable and highly predictive methylation signatures of colorectal cancer and offers a transferable baseline for future large-scale cancer epigenomics studies. Our code is publicly available at https://github.com/matekum/tcga-coad-methylation-ekum-2025a . Author summary: We present a reproducible framework for analyzing genome-wide DNA methylation in colorectal cancer using public TCGA data. Our aim is twofold. We started by making single-site and regional findings statistically reliable by accounting for spatial correlation, hidden confounding, and cell-type mixtures; and second, we evaluate practical classifiers for tumour vs. normal prediction. We smooth effect sizes across neighbouring CpG sites and use permutation tests to detect differentially methylated regions, while surrogate variable analysis and reference-based cell-fraction estimates reduce unwanted variation. We then compare logistic regression with random forests, gradient boosting, and a neural network using principal component features. Across models, we observe near-perfect ranking of samples and strong probability calibration, with logistic regression performing best on calibration. All code, parameters, and figure scripts are openly available so others can reproduce and adapt the framework. Beyond colorectal cancer, the approach provides a template for robust methylome analysis and predictive modelling in other large-scale epigenomic studies.