Raghab Singh
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Raghab Singh has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Pramesh Baral
University of South Dakota
2 shared publications - Akshey Sigdel1 shared publication
- Binamra Neupane
University of Missouri, USA
1 shared publication - Rista Baral1 shared publication
Latest Publications
- Open Access
- 25 April 2026
- Published by AMO Publisher in European Journal of Applied Science, Engineering and Technology
- Vol. 4 (3) , 188-196
Privacy Sentinel: A Multilingual Context-Aware Framework for Real-Time PII Detection and Risk-Based Redaction
The fast rate of development of web content has augmented the risk of privacy violations, specifically, the incidental disclosure of Personally Identifiable Information (PII). To cope with this problem, this paper introduces Privacy Sentinel, a multi-layered system, which is capable of identifying and redacting sensitive information in unstructured and semi-structured web content in real time. The system uses the latest context-sensitive transformer models, such as DeBERTa-v3 and hybrid RoBERTa-LLM models, to detect possible PII leakage, and calculate the risk dynamically depending on the context information. Privacy Sentinel implements a modular design, which is a microservice-based backend based on FastAPI and Redis with a lightweight browser extension on the client. The extension communicates directly with the Document Object Model (DOM), which allows the redaction to be reversible in order to allow sensitive information to be concealed temporarily and still make it available to authorized users when needed. We test the system on big web text corpus and multilingual datasets, where it has a high performance of about 96% accuracy, 94% recall and a F1-score of about 95, which is higher than the traditional regex based systems, single RoBERTa models and the Privacy Shield baseline. The system also has a low latency of approximately 50 milliseconds per DOM element, which ensures that the user can easily switch between redaction on or off without interfering with the page readability. Altogether, our findings indicate that the integration of sophisticated NLP frameworks with the workable system design can offer an efficient and convenient solution to the security of confidential data in contemporary internet-based settings.
- 25 April 2026
- Published by AMO Publisher in European Journal of Applied Science, Engineering and Technology
- Vol. 4 (3) , 188-196
Open AccessPrivacy Sentinel: A Multilingual Context-Aware Framework for Real-Time PII Detection and Risk-Based Redaction
The fast rate of development of web content has augmented the risk of privacy violations, specifically, the incidental disclosure of Personally Identifiable Information (PII). To cope with this problem, this paper introduces Privacy Sentinel, a multi-layered system, which is capable of identifying and redacting sensitive information in unstructured and semi-structured web content in real time. The system uses the latest context-sensitive transformer models, such as DeBERTa-v3 and hybrid RoBERTa-LLM models, to detect possible PII leakage, and calculate the risk dynamically depending on the context information. Privacy Sentinel implements a modular design, which is a microservice-based backend based on FastAPI and Redis with a lightweight browser extension on the client. The extension communicates directly with the Document Object Model (DOM), which allows the redaction to be reversible in order to allow sensitive information to be concealed temporarily and still make it available to authorized users when needed. We test the system on big web text corpus and multilingual datasets, where it has a high performance of about 96% accuracy, 94% recall and a F1-score of about 95, which is higher than the traditional regex based systems, single RoBERTa models and the Privacy Shield baseline. The system also has a low latency of approximately 50 milliseconds per DOM element, which ensures that the user can easily switch between redaction on or off without interfering with the page readability. Altogether, our findings indicate that the integration of sophisticated NLP frameworks with the workable system design can offer an efficient and convenient solution to the security of confidential data in contemporary internet-based settings.
- Open Access
- 25 April 2026
- Published by AMO Publisher in European Journal of Applied Science, Engineering and Technology
- Vol. 4 (3) , 48-57
A Machine Learning Framework for Detecting and Mitigating Adversarial Prompt Attacks in Large Language Models
Large Language Models LLMs have changed numerous industries including education cybersecurity healthcare robotics and Internet of Things IoT by empowering them to converse and comprehend the context and generate intelligent arguments to power intelligent systems. Since these models are deployed as part of software platforms robotics and linked infrastructures their reliability and trustworthiness becomes a key issue. Nonetheless, the bulk of the available literature is concerned with the optimization of model performances and its usage and security risks and adversarial threats are not given much consideration. New attacks like jailbreaking can bypass safety nets and cause great harm to users and systems based on LLMs like multi turn manipulation and injection by emergency attacks. This paper fills this gap by building and analyzing a varied dataset of prompt and safe and adversarial prompts generated by human professionals and advanced LLMs including GPT4o and Claude 3.5 Sonnet. The controversial intent is detected by use of TF IDF feature extraction and machine learning classifiers such as K Nearest Neighbors Decision Tree Logistic Regression random forest Naive Bayes Support Vector machine and ensemble model. The SVM model is the most successful model with 98.4 percent accuracy and 98.7 percent F1 score as well as 99.7 percent recall.
- 25 April 2026
- Published by AMO Publisher in European Journal of Applied Science, Engineering and Technology
- Vol. 4 (3) , 48-57
Open AccessA Machine Learning Framework for Detecting and Mitigating Adversarial Prompt Attacks in Large Language Models
Large Language Models LLMs have changed numerous industries including education cybersecurity healthcare robotics and Internet of Things IoT by empowering them to converse and comprehend the context and generate intelligent arguments to power intelligent systems. Since these models are deployed as part of software platforms robotics and linked infrastructures their reliability and trustworthiness becomes a key issue. Nonetheless, the bulk of the available literature is concerned with the optimization of model performances and its usage and security risks and adversarial threats are not given much consideration. New attacks like jailbreaking can bypass safety nets and cause great harm to users and systems based on LLMs like multi turn manipulation and injection by emergency attacks. This paper fills this gap by building and analyzing a varied dataset of prompt and safe and adversarial prompts generated by human professionals and advanced LLMs including GPT4o and Claude 3.5 Sonnet. The controversial intent is detected by use of TF IDF feature extraction and machine learning classifiers such as K Nearest Neighbors Decision Tree Logistic Regression random forest Naive Bayes Support Vector machine and ensemble model. The SVM model is the most successful model with 98.4 percent accuracy and 98.7 percent F1 score as well as 99.7 percent recall.
- Open Access
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Generative AI for 3D Molecular Structure Prediction Using Diffusion Models
It is essential to be able to create unbiased biological sub-comparative routes to study, to gain an empirical understanding of them. This study explains the exploration of a framework that uses generative artificial intelligence and diffusion-based learning. When drawing graph-based representations having geometric limitations the structures and atom to atom interactions must be valid. Enhancement in molecular stability accuracy achieved on the experimental assessment relative to customary generative approaches. These discoveries work against drugs design and discoveries using AI-driven methods are advantageous.
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessGenerative AI for 3D Molecular Structure Prediction Using Diffusion Models
It is essential to be able to create unbiased biological sub-comparative routes to study, to gain an empirical understanding of them. This study explains the exploration of a framework that uses generative artificial intelligence and diffusion-based learning. When drawing graph-based representations having geometric limitations the structures and atom to atom interactions must be valid. Enhancement in molecular stability accuracy achieved on the experimental assessment relative to customary generative approaches. These discoveries work against drugs design and discoveries using AI-driven methods are advantageous.
- Open Access
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Microarchitectural Advancements in Branch Prediction: Evaluating Modern Techniques for Out-of-Order RISC-V Processors
Today’s out-of-order processors include branch prediction in their microarchitecture to improve the throughput, execution and performance. The aim of this research paper is to evaluate the advanced branch prediction techniques including TAGE-SC-L, L-TAGE, BiMODE, Tournament Predictors, and Multi-Perspective Perceptrons throughout outof-order RISC-V processor design. We use SPEC CPU 2017 benchmark suite in the gem5 simulation framework to study the accuracy of predictions, mispredictions and execution time for different workloads. One of the first uniform simulation-based studies of traditional and MLbased predictors for RISC-V which illustrates practical trade-offs between power and hardware complexity is the current work. Findings suggest TAGE-SC-L provides an optimal trade-off in accuracy and efficiency enabling improved prediction strategies in systems that are energy-wasteful for best outcomes. Findings can influence the microarchitecture choices of future RISC-V implementations.
- 27 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessMicroarchitectural Advancements in Branch Prediction: Evaluating Modern Techniques for Out-of-Order RISC-V Processors
Today’s out-of-order processors include branch prediction in their microarchitecture to improve the throughput, execution and performance. The aim of this research paper is to evaluate the advanced branch prediction techniques including TAGE-SC-L, L-TAGE, BiMODE, Tournament Predictors, and Multi-Perspective Perceptrons throughout outof-order RISC-V processor design. We use SPEC CPU 2017 benchmark suite in the gem5 simulation framework to study the accuracy of predictions, mispredictions and execution time for different workloads. One of the first uniform simulation-based studies of traditional and MLbased predictors for RISC-V which illustrates practical trade-offs between power and hardware complexity is the current work. Findings suggest TAGE-SC-L provides an optimal trade-off in accuracy and efficiency enabling improved prediction strategies in systems that are energy-wasteful for best outcomes. Findings can influence the microarchitecture choices of future RISC-V implementations.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Reinforcement Learning for Robotic Singulation: Policy Space Impact on Interactive Manipulation
This paper, entitled Reinforcement Learning for Robotic Singulation: Policy Space Impact on Interactive Manipulation, investigates how the design of policy space influences reinforcement learning (RL) performance in robotic manipulation. Specifically, we compare Proximal Policy Optimization (PPO) with continuous action space against Deep Q-Learning (DQL) with discretized action space in a simulated singulation environment. The objective is to analyze how action space complexity affects convergence speed, policy stability, and manipulation success. Empirical findings show that while PPO policies capture fine-grained motions, they suffer in multimodal reward landscapes, whereas DQL demonstrates higher robustness, faster convergence, and consistent success rates in cluttered scenarios. This study highlights the critical role of action space design in RL-based robotic singulation, providing actionable insights for robust and scalable real-world deployment.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessReinforcement Learning for Robotic Singulation: Policy Space Impact on Interactive Manipulation
This paper, entitled Reinforcement Learning for Robotic Singulation: Policy Space Impact on Interactive Manipulation, investigates how the design of policy space influences reinforcement learning (RL) performance in robotic manipulation. Specifically, we compare Proximal Policy Optimization (PPO) with continuous action space against Deep Q-Learning (DQL) with discretized action space in a simulated singulation environment. The objective is to analyze how action space complexity affects convergence speed, policy stability, and manipulation success. Empirical findings show that while PPO policies capture fine-grained motions, they suffer in multimodal reward landscapes, whereas DQL demonstrates higher robustness, faster convergence, and consistent success rates in cluttered scenarios. This study highlights the critical role of action space design in RL-based robotic singulation, providing actionable insights for robust and scalable real-world deployment.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Next-Gen Satellite Imaging: Optimizing High-Resolution Remote Sensing with Advanced Stabilization and Data Processing
The evolution of satellite imaging is driven by the need for high-resolution, cost-effective, and data-efficient remote sensing systems. This paper explores novel imaging modalities and hybrid stabilization techniques that enhance the trade-offs between image quality, bandwidth, and system size. By leveraging high-speed CMOS sensors, digital oversampling, and onboard processing, we propose a framework that improves data efficiency, spectral imaging, and dynamic range for next-generation small satellite constellations. Our analysis demonstrates how adaptive sensor technologies and computational imaging can redefine the landscape of high-resolution Earth observation while maintaining scalability and affordability.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessNext-Gen Satellite Imaging: Optimizing High-Resolution Remote Sensing with Advanced Stabilization and Data Processing
The evolution of satellite imaging is driven by the need for high-resolution, cost-effective, and data-efficient remote sensing systems. This paper explores novel imaging modalities and hybrid stabilization techniques that enhance the trade-offs between image quality, bandwidth, and system size. By leveraging high-speed CMOS sensors, digital oversampling, and onboard processing, we propose a framework that improves data efficiency, spectral imaging, and dynamic range for next-generation small satellite constellations. Our analysis demonstrates how adaptive sensor technologies and computational imaging can redefine the landscape of high-resolution Earth observation while maintaining scalability and affordability.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Exploring Exhaustivity in Wh-Questions through Analysis of Natural Language Usage
This study investigates the psychological factors that influence how a hearer interprets a speaker's intention in response to a question. The analysis focuses on wh-questions and the extent of exhaustivity required to fulfill the speaker's communicative goal. While previous research suggests that such questions typically elicit exhaustive answers, the findings presented here indicate that exhaustivity is not an inherent property of the question itself. Question paraphrase ratings were collected across varied linguistic contexts to support this conclusion. The results highlight the significance of both linguistic and discourse-level considerations-particularly the hearer's assessment of the speaker's intent-in shaping the interpretation of meaning in question-answer exchanges.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessExploring Exhaustivity in Wh-Questions through Analysis of Natural Language Usage
This study investigates the psychological factors that influence how a hearer interprets a speaker's intention in response to a question. The analysis focuses on wh-questions and the extent of exhaustivity required to fulfill the speaker's communicative goal. While previous research suggests that such questions typically elicit exhaustive answers, the findings presented here indicate that exhaustivity is not an inherent property of the question itself. Question paraphrase ratings were collected across varied linguistic contexts to support this conclusion. The results highlight the significance of both linguistic and discourse-level considerations-particularly the hearer's assessment of the speaker's intent-in shaping the interpretation of meaning in question-answer exchanges.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Gesture Recognition for Remote Control Systems Using Kernel Principal Component Analysis (KPCA) and Trajectory Normalization
This paper presents an integrated framework that combines Kernel Principal Component Analysis (KPCA) with trajectory normalization for robust classification of hand gestures in non-contact remote control systems. The approach supports both mouse-tracked and visionbased gesture data, enabling source-agnostic recognition. Trajectory normalization is employed to standardize input variations caused by gesture speed, orientation, and sampling inconsistencies, while KPCA effectively captures non-linear gesture patterns in reduced dimensions. The framework incorporates and evaluates various interpolation techniques to refine temporal alignment across input sequences. Experimental results demonstrate improved accuracy in recognizing complex and degenerate gesture patterns, particularly in scenarios involving user inconsistency and noisy input. Furthermore, the paper provides a comprehensive analysis of the trade-offs involved in different normalization schemes and their impact on classifier performance. This work contributes to the development of intuitive and reliable human-computer interaction systems, suitable for smart environments and assistive technologies. Experimental results show over 99.9% accuracy for non-degenerate gestures and over 98% for degenerate gestures using distance-based normalization with uniform scaling. The system demonstrates strong generalization across input modalities. This work paves the way for real-world contactless control in assistive and industrial domains.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessGesture Recognition for Remote Control Systems Using Kernel Principal Component Analysis (KPCA) and Trajectory Normalization
This paper presents an integrated framework that combines Kernel Principal Component Analysis (KPCA) with trajectory normalization for robust classification of hand gestures in non-contact remote control systems. The approach supports both mouse-tracked and visionbased gesture data, enabling source-agnostic recognition. Trajectory normalization is employed to standardize input variations caused by gesture speed, orientation, and sampling inconsistencies, while KPCA effectively captures non-linear gesture patterns in reduced dimensions. The framework incorporates and evaluates various interpolation techniques to refine temporal alignment across input sequences. Experimental results demonstrate improved accuracy in recognizing complex and degenerate gesture patterns, particularly in scenarios involving user inconsistency and noisy input. Furthermore, the paper provides a comprehensive analysis of the trade-offs involved in different normalization schemes and their impact on classifier performance. This work contributes to the development of intuitive and reliable human-computer interaction systems, suitable for smart environments and assistive technologies. Experimental results show over 99.9% accuracy for non-degenerate gestures and over 98% for degenerate gestures using distance-based normalization with uniform scaling. The system demonstrates strong generalization across input modalities. This work paves the way for real-world contactless control in assistive and industrial domains.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Evolutionary AI for Morphological and Locomotor Design of Resilient Exploration Robots
We present an AI-driven evolutionary framework that co-designs morphology and locomotion for compliant (soft) robots intended for extraterrestrial terrains. Using CPPN-NEAT with novelty-guided search, our approach explores diverse body plans and gaits across Earth, Martian, and lunar gravity. In an ablation against direct encodings, CPPN-based robots achieved 1.78× greater mean displacement under Earth gravity (14.3 cm vs. 8.0 cm), while also exhibiting higher morphology-diversity scores (0.87 vs. 0.41), indicating improved exploration of the design space. Novelty search further enriched behavior-space coverage and yielded robust gaits under low gravity. Results highlight that diversity-centric, generative encodings can surface unconventional yet effective designs often overlooked by purely fitness-driven optimization.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessEvolutionary AI for Morphological and Locomotor Design of Resilient Exploration Robots
We present an AI-driven evolutionary framework that co-designs morphology and locomotion for compliant (soft) robots intended for extraterrestrial terrains. Using CPPN-NEAT with novelty-guided search, our approach explores diverse body plans and gaits across Earth, Martian, and lunar gravity. In an ablation against direct encodings, CPPN-based robots achieved 1.78× greater mean displacement under Earth gravity (14.3 cm vs. 8.0 cm), while also exhibiting higher morphology-diversity scores (0.87 vs. 0.41), indicating improved exploration of the design space. Novelty search further enriched behavior-space coverage and yielded robust gaits under low gravity. Results highlight that diversity-centric, generative encodings can surface unconventional yet effective designs often overlooked by purely fitness-driven optimization.
- Open Access
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Neural Network-Based Surrogate Modeling for Fast Solid Mechanics Simulations
Solutions of solid mechanics problems are available with reasonable accuracy through finite element analyses. However, performance studies with larger magnitude and real-time applications aren’t computationally viable. This paper proposes a compact neural network surrogate framework that easily learns the material and geometric parameters to nodal displacement mapping within the set-up a Multi-Layer Perceptron (MLP). The datasets from the ONSAS.m solver are parameter sweeps of two benchmark problems (a uniaxial prism and a composite cantilever). Our surrogate trained with domain RMSE is lower and it is significantly faster by orders of magnitude. We also verify extrapolation performance and visually compare displacements produced by the model with the ones from FEM. To sum up, we provide suggestions for consistency and uncertainty quantification for physics through operator learning and physicsinformed models.
- 26 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessNeural Network-Based Surrogate Modeling for Fast Solid Mechanics Simulations
Solutions of solid mechanics problems are available with reasonable accuracy through finite element analyses. However, performance studies with larger magnitude and real-time applications aren’t computationally viable. This paper proposes a compact neural network surrogate framework that easily learns the material and geometric parameters to nodal displacement mapping within the set-up a Multi-Layer Perceptron (MLP). The datasets from the ONSAS.m solver are parameter sweeps of two benchmark problems (a uniaxial prism and a composite cantilever). Our surrogate trained with domain RMSE is lower and it is significantly faster by orders of magnitude. We also verify extrapolation performance and visually compare displacements produced by the model with the ones from FEM. To sum up, we provide suggestions for consistency and uncertainty quantification for physics through operator learning and physicsinformed models.