International Journal of Ad Hoc and Ubiquitous Computing (20 papers in press)

Regular Issues

• Analysing Theoretical and Neural Distinguishers in Salsa 128 Bits  
  by Karthika SK, Kunwar Singh 
  Abstract: One of the finalists for the eSTREAM projects in 2005 was Salsa, created by Daniel J. Bernstein. Salsa is a widely recognised stream cipher that gained prominence after multiple cryptanalytic techniques were applied to the popular stream cipher RC4. Salsa offers two variations, with key sizes of 128 bits and 256 bits, depending on the seed key’s length. Salsa has undergone multiple key recovery attacks, particularly targeting the 256-bit variant, reaching up to its eighth round. Additionally, numerous experimental attacks have been conducted on Salsa, leaving room for further theoretical analysis. Theoretical analysis plays a crucial role in identifying vulnerable aspects of the cipher, enabling the design of stronger ciphers that are resistant to attacks. In a study by Dey and Sarkar (2021), they conducted a theoretical analysis to examine the origins of distinguishers found in experimental attacks on Salsa 256 bits and Chacha 256 bits. Inspired by their work, our paper focuses on the theoretical analysis of a differential attack on Salsa 128 bits, specifically up to four rounds. We mathematically established the probabilities of various observations and our theoretical analysis aligns with the experimental findings for Salsa 128 bits. Further, we have also found neural distinguisher for Salsa 128 bits based on the model proposed by Gohr (2019) in CRYPTO 2019. The designed model has found neural distinguishers at round 4 with 72.1% accuracy.
  Keywords: stream cipher; Salsa; cryptanalysis; differential attack; theoretical analysis; deep learning.
  DOI: 10.1504/IJAHUC.2024.10068235
   
  
• Improving IoT Access Control with Context-Aware Machine Learning: Reducing Bias and Enhancing Accuracy  
  by Surendra Tyagi, Yamuna Prasad, Devesh C. Jinwala, Subhasis Bhattacharjee 
  Abstract: With the vast amounts of data generated in typical IoT applications, there are challenges in collecting, modelling, reasoning, and distributing the context of the sensed data. The context of the sensor data can trigger an idea about using it in many applications where access can be regulated partially or entirely. Traditional access control methods provide coarse controls, viz., full or no access. Fine-grained access control can be developed using context awareness. Policy-based access control methods in dynamic environments require periodic updates to adapt policies based on data patterns. Machine learning can be used to an advantage when an access control method requires periodic updates of policies in dynamic environments, focusing on generating and adapting policies based on data patterns. This paper investigates applying context-aware machine learning (CAML) models, which reduces bias in learning while simultaneously building policies based on patterns inferred from trustworthy access reports available as past data. CAML improved accuracy to 99.9% for the smart home dataset in most cases.
  Keywords: Context-aware; Machine Learning; Internet of Things; Access Control.
  DOI: 10.1504/IJAHUC.2024.10068368
   
  
• Know Your Device (KYD): a Blockchain-Based Self-sovereign Identity Management Framework for IoT Devices  
  by Aswani Aguru, Suresh Babu E 
  Abstract: The internet of things (IoT) comprises billions of devices that communicate and exchange data through the internet. However, managing each devices registration, identification, and authentication information is challenging. In this paper, we have leveraged the efficient management of the self-sovereign identity of IoT devices using the blockchain network called know your device. The blockchain validators act as identity-verifiers for the authenticated IoT devices through threshold-based signatures. The verifiers are elected in each tenure by a lightweight proof-of-voting consensus algorithm. As the off-chain mechanism, the interplanetary file system (IPFS) stores the registered IoT devices identities and authentication credentials. We have performed the experimental result analysis of the proposed scheme on Hyperledger Fabric. Privacy, anonymity, and accountability are significant achievements of the proposed scheme. The results have proven the efficiency of our scheme in terms of lower execution time, lower smart contract deployment time, and higher transaction throughput than the state-of-the-art techniques.
  Keywords: Internet of Things; Self-sovereign Identity; Threshold Signatures; Authentication; Blockchain; Proof-of-Voting Consensus; Privacy; Anonymity; Accountability.
  DOI: 10.1504/IJAHUC.2024.10068514
   
  
• An Energy Harvesting Fairness Maximisation for Active/Passive IRS-assisted PS-SWIPT System  
  by Tuan Pham, Van-Quang-Binh Ngo, Pham Ngoc Son, Van Hiep VU 
  Abstract: In this study, broadcast power-splitting simultaneous wireless information and power transfer (SWIPT) assisted by active/passive intelligent reflecting surface (IRS) is investigated to enhance the harvested energy of receivers under the constraints of data rate and limited transmission powers. The maximization of minimum harvested energy is investigated by jointly optimising the beamforming vector and power-splitting (PS) factors at the transmitter and the receivers, the amplification factors, and/or the phase shifts at the active/passive IRS. The alternating optimisation, successive convex approximations, and feasible point pursuit methods cooperate to find the near-optimal solution. Finally, the numerical simulation presents the convergence and effectiveness of the proposed algorithm.
  Keywords: active/passive intelligent reflecting surface; simultaneous wireless information and power transfer; power-splitting; alternating optimisation.
  DOI: 10.1504/IJAHUC.2024.10068599
   
  
• Hybrid Cheetah and Artificial Rabbits Optimisation Algorithm for Micro Electro Mechanical System  
  by Sabitha Balasubramanian, Raffik Rasheed 
  Abstract: In this manuscript, a Hybrid Cheetah and Artificial Rabbits Optimization Algorithm fostered Digital Control Systems for Micro Electro Mechanical Systems Gyroscope(MEMS-GYS-HCO-ARO) is proposed. In this method the design Procedure starts by designing a closed loop control systems for Micro Electro Mechanical Systems Gyroscope and loop parameters are optimized using Hybrid algorithms. First the loop parameters are optimised by the help of HCO and ARO. Then Dwarf Mongoose Optimisation Algorithm and LMS demodulator utilized for demodulate noise signal. The performance of the proposed MEMS-GYS-HCO-ARO method attains amplitude analysis of 39.25%, 33.36% and 32.99% of lower amplitude compared with existing methods, like a design of micro electro mechanical systems gyroscope using Hybrid Optimisation Algorithm (MEMS-GYS-HOA), Digital Control and Readout of MEMS Gyroscope Using Second-Order Sliding Mode Control(MEMS-GYS-SOSMC) and A Digital Closed Loop Sense MEMS Disk Resonator Gyroscope Circuit Design Based on Integrated Analog Front-end (DCLS-MEMS-IAF) respectively.
  Keywords: MEMS gyroscope; closed-loop system; control systems; hybrid cheetah and Artificial Rabbits algorithm; Monte Carlo analysis; demodulator.
  DOI: 10.1504/IJAHUC.2024.10068845
   
  
• A Robust Pipelined Granular Approach for Enhanced Efficiency in Document Text Summarisation Using Hierarchical Categorisation and Feature Extraction Techniques  
  by Krishna Dheeravath, S.Jessica Saritha 
  Abstract: Mixed script inquiries various linguistic phrases make intention identification difficult. This study proposes a new method that uses knowledge-based approaches and RNNs to reliably identify mixed script query intentions. The method uses knowledge-based databases of negative and positive terms to describe people and words that change intention. Language nuances are trained using these datasets. Mixed script queries are pre-processed and translated into numerical vector representations using knowledge-based intent areas. These vector representations are processed using RNN architecture to learn complex input-intention mappings. An attention method can emphasise key inquiries to help identify intentions. A diversified dataset of mixed script queries is used to test the suggested approach using accuracy, precision, recall, and F1-score. The results show the model can reliably discern intentions across languages. By solving mixed script query problems and offering a solid framework for multilingual intention recognition, this research advances natural language processing. The method could be used in information retrieval, sentiment analysis, and cross-language communication.
  Keywords: Feature Extraction; Text Summarisation; Hierarchical Categorisation; Natural Language Processing; Large-scale Document Analytics.
  DOI: 10.1504/IJAHUC.2024.10068897
   
  
• Enhancing Intrusion Detection: Combining Logitboost Algorithms and Random Forest  
  by Ankit Kharwar, Diya Vadhwani, Dipak Dabhi, Vivaksha Jariwala 
  Abstract: Network data security is an issue that affects individuals, businesses, and governments worldwide. As attacks become more common and attackers' tactics evolve, it is important to implement advanced network security solutions such as an intrusion detection system (IDS) to detect unwanted and unexpected network activity. To that end, this article proposes a comprehensive strategy for improving detection performance through classification approaches. If only one classifier is utilised, the final decision may be erroneous, as incorrect classifier output may occur. The ensemble classification method combines multiple classifiers and produces better results than a single classifier. To improve classification accuracy, the proposed model incorporates Random Forest and Logitboost. The proposed model has an accuracy of 95.89%, 99.91%, and 98.54% on the UNSW-NB15, CICIDS2017, and CICIDS2018 datasets, respectively, and outperforms other existing models in terms of accuracy, detection rate, and false alarm rate.
  Keywords: Logitboost Algorithm; Network Security; Anomaly Detection; Machine Learning; Intrusion Detection; Random Forest; Boosting Algorithm; Ensemble methods.
  DOI: 10.1504/IJAHUC.2024.10069005
   
  
• Blockchain Enabled Vehicle NFT and Key Management for IoV Networks in Fog Computing Environment  
  by Brijesh Chaurasia, Vinay Rishiwal, Mano Yadav, Man Mohan Shukla, Preeti Yadav 
  Abstract: In the Internet of Vehicles Networks (IoV-N), vehicles exchange messages to improve traffic efficiency using fog and edge computing. Vehicles may make decisions based on exchanged messages in IoV-N. Therefore, security and key management in the IoV-N are the primary concerns. In this paper, key management for social IoV-N using Blockchain is presented. This work also utilizes non-fungible tokens (NFTs) and interplanetary file systems (IPFS) to store and secure records such as keys, issuer identities, etc., on behalf of trusted static authorities such as the city-level transport authority (CTA). The paper also introduced bilinear pairing for key generation and secure exchange in the IoV-N. CTA makes up the key generation process at the fog layer to reduce latency, and the Blockchain is stored in a data server at the cloud layer. Extensive simulation and results show that key management for social IoV-N is a secure, fast, and viable solution.
  Keywords: Blockchain; Internet of Vehicles (IoV); Fog computing; Cloud computing; NFTs; Interplanetary File System (IPFS).
  DOI: 10.1504/IJAHUC.2024.10069167
   
  
• A Novel Multigrade Classification in FL using Brain MRI Images based on FHAT_EfficientNet  
  by Madan Lal Saini, Aravapalli Rama Satish, Madhusudhana Rao T. V., Jyothi Mandala, Smritilekha Das, Cristin R. 
  Abstract: This paper establishes the Fractional Harmony Artificial Tree (FHAT)_EfficientNet for multi-grade classification in Federated Learning (FL). Here, the established FHAT is attained by the integration of the Fractional Calculus (FC) and Harmony search-based Feedback Artificial Tree (HSFAT) algorithm, and the HSFAT is developed by the combination of Harmony Search (HS) and Feedback Artificial Tree (FAT). Initially, the input MRI image is taken from a particular dataset and subjected to pre-processing. Thereafter, tumor segmentation is accomplished based on Fuzzy Local Information C-Means (FLICM). Later, image augmentation and feature extraction are performed. Finally, the multi-grade classification is carried out using EfficientNet fine-tuned based on the proposed FHAT. Moreover, the established FHAT_EfficientNet attained better accuracy, specificity, sensitivity, Mean Squared Error (MSE), Root Mean Square Error (RMSE), and loss function of 0.917, 0.936, 0.966, 0.058, 0.241, and 0.083.
  Keywords: EfficientNet; Fuzzy Local Information C-Means; Federated Learning; Fractional Harmony Artificial Tree; and Fractional Calculus.
  DOI: 10.1504/IJAHUC.2024.10069264
   
  
• Hybrid Fuzzy Based Shepard Convolutional Maxout Network- Based Skin Cancer detection  
  by Pakutharivu P, Santhi K, Chellatamilan T, Ramanathan Lakshmanan 
  Abstract: Skin cancer arises from the uncontrolled growth of skin cells. Timely detection can significantly lower mortality rates and improve patient outcomes. However, diagnosing melanoma can be challenging due to its similarity to benign lesions. This study introduces a Fuzzy Shepard Convolutional Neural Network (FSCMN) for detecting skin cancer in images. Initially, the pre-processing is done by using a bilateral filter. Then, the skin lesions are segmented by using Recurrent Prototypical Network (RP-Net). Next, features are extracted using Convolutional Neural Networks (CNN), Local Vector Pattern (LVP), Gray Level Co-occurrence Matrix (GLCM) texture features, and Entropy-based Local Directional Texture Pattern (LDTP). Finally, skin cancer detection is performed by proposed FSCMN, which integrates Fuzzy Logic, Shepard Convolutional Neural Network (ShCNN), and Deep Maxout Network (DMN). The FSCMN approach achieved impressive results, with accuracy, True Positive Rate (TPR), and True Negative Rate (TNR) of 0.928, 0.939, and 0.914.
  Keywords: Fuzzy Concept; Recurrent Prototypical Network; Local Vector Pattern; Shepard Convolutional Neural Network; Deep Maxout Network.
  DOI: 10.1504/IJAHUC.2024.10069439
   
  
• Transformer-based Infrared and Visible Image Fusion with Language-driven Loss  
  by Qihao Liu, Fei Lu, Ningxin Wang, Jungan Zhang, Yuanchao Hu 
  Abstract: Infrared and visible image fusion aims to generate a fused image that highlights the target and encompasses detailed textural information. AI-driven image fusion enables enhanced multimodal sensing and decision-making across diverse environments. However, existing algorithms often focus only on visual quality while neglecting the image's semantic content. To address these issues, we propose an image fusion method that leverages the global feature extraction capabilities of the transformer model and optimizes the fusion process through a loss function guided by the Contrastive Language-Image Pre-training~(CLIP). Firstly, we develop a feature-guided transformer~(FGT) module to extract and interact with local and global information in images. Subsequently, a feature dynamic fusion~(FDF) module is designed to fuse the image adaptively from two different modalities. Comprehensive experiments conducted on three public datasets demonstrate that the proposed method outperforms state-of-the-art (SOTA) fusion methods subjectively and objectively.
  Keywords: Image Fusion; Language-Vision Pre-training; Transformer; IoT Network.
  DOI: 10.1504/IJAHUC.2025.10069553
   
  
• An Efficient Approach for Congestion Detection and Control in VANETs  
  by Brijesh Chaurasia, Vinay Rishiwal, Mohammad Shiblee, Man Mohan Shukla, Mano Yadav 
  Abstract: Vehicular Ad-hoc Network (VANET) is a promising technology that uses a variety of messages to deliver convenience and safety features. VANET has unique characteristics regarding the high speed and variable density of vehicles. The short time for vehicular communication, especially at high speeds, presents difficulties. Channel utilization, especially for safety messages, is a big issue due to its unique characteristics, which can lead to congestion, higher delays, and energy waste. This problem affects the performance of the VANET as a whole and the efficiency of VANET applications, such as the emergency notification system. This work aims to design a novel mechanism for managing and detecting congestion in VANET to mitigate these issues. The proposed mechanism uses priority-based channel assignment to handle congestion control in VANET. The extensive simulation results demonstrate that the proposed approach effectively addresses congestion, ensuring a more reliable environment for VANET operations.
  Keywords: VANETs; Congestion Control; Message Scheduling; Priority Assignment; Communication.
  DOI: 10.1504/IJAHUC.2024.10069614
   
  
• CNN based Hybrid Precoding Framework using Dual-Phase Shifter in MIMO Systems  
  by Deepti Sharma, Ramesh Babu Battula 
  Abstract: Emerging applications such as autonomous driving and intelligent healthcare demand ultra-low latency and precise positioning for efficient beamforming, essential for advanced 5G technologies. Hybrid beamforming (HB) has gained prominence as a technique to balance hardware complexity with transmission rates in 5G networks. However, traditional HB methods are computationally intensive and underutilize channel state information, leading to reduced spectral efficiency and transmission rates. Deep learning has revolutionized wireless communication by providing optimised solutions to complex challenges. This work proposes a convolutional neural network (CNN)-based HB framework to minimise complexity while achieving optimal beamforming with enhanced transmission rates. The framework incorporates a hybrid precoding network, HP-CNN, featuring a dual-phase analog shifter and a rate optimiser. Simulations demonstrate that the proposed framework surpasses conventional algorithms, offering superior performance and efficiency. By leveraging deep learning, this approach addresses key challenges in HB, paving the way for robust and high-performance 5G communication systems.
  Keywords: Hybrid Beamforming; Convolutional Neural Network (CNN); Dual-Phase Shifter; Channel-State Information (CSI); Multiple-Input-Multpile-Output (MIMO).
  DOI: 10.1504/IJAHUC.2025.10069682
   
  
• Adaptive Optimisation of a Resource Allocation Algorithm for Secure Video Transmission in 5G Networks  
  by Augustin Minalkar, Srinath Doss, Ruchi Doshi 
  Abstract: This paper presents a novel optimisation model Tasmanian Devil Whale Optimization (TDWO) for secure transmission of educational video using Fifth-Generation (5G) cellular networks. At first, the input educational videos taken from the database is transmitted via 5G networks. Then, the allocation of resource is performed using the designed TDWO model by considering different fitness parameters, like data rate, achievable data rate, and Quality of Experience (QoE). Here, the deep learning model, Deep Convolutional Neural Networks (DCNN) is utilised for the prediction of QoE for resource allocation. Moreover, the resource allocation performance of the TDWO is validated by comparing with other resource allocation schemes. Here, the TDWO algorithmic approach achieved superior performance with throughput of 25.557Mbps accuracy of 91.43%, Bit Error Rate (BER) of 0.021, QoE of 18.332, and fitness of 0.013.
  Keywords: Deep Convolutional Neural Networks; Tasmanian Devil Whale Optimization; Whale Optimization Algorithm; Tasmanian Devil Optimization; Network resource allocation.
  DOI: 10.1504/IJAHUC.2024.10069697
   
  
• Big Data Classification using Mapreduce Framework Enabled Deep Quantum Dilated Convolutional Neural Network  
  by Nandini M., Deepak N. Biradar 
  Abstract: In general, big data deals with the examination of enormous amount of data from distributed regions utilising ML modules. A big data classification technique using a hybrid deep learning-based optimisation algorithm in the MapReduce framework is proposed. Initially, data partitioning is conducted using Deep Fuzzy Clustering (DFC). In the mapper phase, data normalisation is performed using linear normalisation. Then, feature fusion is done by Neyman measure with Deep Residual Network (DRN). Later, data augmentation is performed using the oversampling model Synthetic Minority Oversampling Technique (SMOTE). Finally, in the reducer phase, big data classification is done utilizing Deep Quantum Dilated Convolutional Neural Network (DQDCNN) model merged by Quantum Dilated Convolutional Neural Network (QDCNN) and Deep Quantum Neural Network (DQNN). Here, DQDCNN is trained by Gazelle Hunter Optimization (GHO) developed by Gazelle Optimization Algorithm (GOA) and Hunter Prey Optimization (HPO). The measures utilised for GHO-DQDCNN obtained supreme values of 92.4%, 92.9% and 92.6%.
  Keywords: Big data classification; MapReduce framework; Deep Fuzzy Clustering (DFC); Synthetic Minority Oversampling Technique (SMOTE); Deep Quantum Neural Network (DQNN).
  DOI: 10.1504/IJAHUC.2024.10069707
   
  
• Fractional Archimedes Optimisation Algorithm Enabled Deep Learning for Diabetic Retinopathy Detection  
  by Kundan B, Pushpa S. 
  Abstract: This paper proposes a fractional Archimedes optimisation algorithm (FAOA)-based deep learning model for diabetic retinopathy (DR) detection. Here, the FAOA is formed by combining fractional calculus with Archimedes optimisation algorithm (AOA). Initially, in the pre-processing step, the noise from the input image is removed by the Kalman filter and region of interest (RoI) is extracted. Then, optic disc segmentation and blood vessel segmentation are done. After that, the features are extracted from both segmented outputs. Simultaneously, features are taken from the Haar wavelet transform, which is derived from the input image. Then, extracted features are allowed to FAOA_LeNet for DR detection. The performance is analysed using the Indian diabetic retinopathy image dataset and digital retinal images for vessel extraction dataset. The performance obtained by the proposed FAOA_LeNet in terms of sensitivity, accuracy, and specificity is 0.938, 0.954, and 0.974.
  Keywords: Kalman filter; Archimedes Optimization Algorithm; Fuzzy C-Means clustering; Fractional Calculus; Haar Wavelet Transform.
  DOI: 10.1504/IJAHUC.2025.10070019
   
  
• Investigation of pixel neighbourhood for prediction error-based reversible data hiding using neural networks  
  by Sabhapathy Myakal, Rajarshi Pal, Nekuri Naveen 
  Abstract: Success of a prediction error-based reversible data hiding (RDH) technique depends on a good pixel value predictor. A pixel value predictor predicts a pixel value with the help of its neighbouring pixel values. Considered neighbourhood for pixel value prediction influences the performance of such predictor. Varieties of pixel neighbourhoods have been considered in literature. This paper presents a unique work of its kind to explore the effect of various pixel neighbourhoods. In order to perform this comparative analysis, a neural network is used here to predict a pixel value from a neighbourhood. The best neighbourhood for the pixel value prediction task is determined from the reported experimental observations. The selected neighbourhood is used for pixel value prediction and subsequent RDH scheme. It is observed from the experimental results that the proposed neural network-based adaptive RDH scheme with the selected neighbourhood outperforms majority of the state-of-the-art RDH techniques.
  Keywords: reversible data hiding; RDH; pixel neighbourhood; pixel value prediction; prediction error; multi-layer perceptron; MLP.
  DOI: 10.1504/IJAHUC.2024.10067149
   
  
• A three-stage approach using deep learning for automated vehicle smart parking with license plate recognition  
  by Shaunak Gupta, Pushkar Garg, Abhinav Aggarwal, Gaurav Goyal, Kanu Goel 
  Abstract: Urbanisation and increasing vehicular density have amplified challenges in conventional parking systems, necessitating innovative solutions for effective parking management. Common challenges include the absence of real-time information on available parking slots, limited data analysis tools for space utilisation, reliance on manual processes, and the inadequate incorporation of automation at entry and exit of vehicles. The proposed smart parking approach endeavours to mitigate these challenges by integrating advanced technologies to enable efficient parking slot detection, comprehensive data analysis, and seamless automation. It integrates automation at entry and exit points, streamlining the parking process. Leveraging advanced computer vision and sensor technologies, the system provides real-time identification of empty parking slots, enhancing overall space utilisation. The system also incorporates accident tracking mechanisms to enhance safety within parking facilities. The research paper also presents a novel accident detection model attaining a commendable accuracy of 94.5%.
  Keywords: artificial intelligence; smart parking; deep learning; automated number plate recognition; ANPR; pre-trained models; convolutional neural network; CNN.
  DOI: 10.1504/IJAHUC.2024.10067924
   
  
• Performance analysis of the contention-based access periods of the IEEE 802.11ad hybrid MAC based on Pareto arrivals  
  by R.G. Sangeetha, C. Hemanth, T.G. Venkatesh 
  Abstract: IEEE 802.11ad is a wireless LAN standard that operates in the 60 GHz range. In this article, we undertake a simulation-based performance evaluation of the IEEE 802.11ad, hybrid MAC protocol's contention-based access periods (CBAPs) under Pareto arrivals. In this article we evaluate the performance in terms of throughput and average latency. Because of the protocol's hybrid nature, CBAP packet transfers must be deferred when the protocol transitions from CBAP to a contention-free phase. We demonstrate that: 1) deferring packet transmission; 2) the CBAP bandwidth allocation method have a substantial impact on the system performance. We also highlight the difference in performance that is achieved when a heavy-tailed traffic model like Pareto is used as against the traditional Poisson model.
  Keywords: hybrid MAC; IEEE 802.11ad; Pareto; throughput; MAC delay; simulation.
  DOI: 10.1504/IJAHUC.2024.10067923
   
  
• CSPO: chain-structured prompt optimisation for large language models  
  by Jinshui Wang, Sining Lin, Xingsi Xue, Shuguang Chen, Zhengyi Tang 
  Abstract: Large language models (LLMs) show promise in improving content distribution in mobile communication networks, but their performance heavily depends on input prompts. Manually crafting effective prompts for complex tasks is time-consuming and often suboptimal, highlighting the need for automated optimisation. This paper proposes a chain-structured prompt optimisation (CSPO) method to automatically optimise prompts for LLMs. Inspired by neural network training, CSPO decomposes tasks into a series of ordered instruction steps, forming an instruction chain. It then employs an iterative optimisation process analogous to forward and backward propagation to refine this chain. CSPO automatically constructs an initial instruction chain, executes it to generate output, analyses errors, formulates improvement strategies, and updates the chain accordingly. Evaluations on instruction induction, mathematical reasoning, and counter factual tasks demonstrate that CSPO generally outperforms baseline methods. This study contributes to the field of prompt engineering, offering a novel approach to automatic prompt optimisation for LLMs.
  Keywords: large language models; automatic prompt optimisation; natural language processing; mobile communication networks; content distribution.
  DOI: 10.1504/IJAHUC.2025.10068590