Praveen Kumar Myakala
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Praveen Kumar Myakala has a total of 16 co-authors. Here are the top co-authors by number of shared publications:
- Anil Kumar Jonnalagadda
Independent Researcher,Frisco,TX,USA,75035
5 shared publications - Chiranjeevi Bura
University of Colorado Boulder, Boulder, CO 80309, United States
5 shared publications - Srikanth Kamatala
Independent Researcher,Texas,USA
3 shared publications Fnu Samaah
2 shared publications- Direesh Reddy Aunugu
Independent Researcher
1 shared publication - Madhura Raut
University of Southern California
1 shared publication
Latest Publications
- 11 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 553-559
Fine-Tuned Vs Prompted Models for Emotion Detection: A Robustness and Slice-Based Study
Fine-grained emotion detection on social media is difficult due to informal language and stylistic noise. This paper compares (i) a supervised fine-tuned BERT classifier and (ii) instruction-tuned large language models (Mistral, LLaMA, Gemma) used in zero-shot prompting on a single-label subset of GoEmotions. We report Accuracy, Macro-F1, and WeightedF1, and extend standard evaluation with robustness tests under realistic perturbations (typos, token dropout, punctuation noise, hashtag transformations). We also perform slice-based diagnostics using heuristic linguistic markers (emoji use, capitalization, pronouns, profanity, negation) to surface performance disparities that aggregate metrics can hide. Results show that fine-tuned BERT performs best on clean data, while prompted LLMs are generally more tolerant to surface noise but have lower absolute accuracy. Slice analysis highlights consistent degradation on profanity and negation, indicating persistent challenges on informal, style-heavy text. These findings provide practical guidance for deploying emotion models in noisy social media settings.
- 11 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 553-559
Fine-Tuned Vs Prompted Models for Emotion Detection: A Robustness and Slice-Based Study
Fine-grained emotion detection on social media is difficult due to informal language and stylistic noise. This paper compares (i) a supervised fine-tuned BERT classifier and (ii) instruction-tuned large language models (Mistral, LLaMA, Gemma) used in zero-shot prompting on a single-label subset of GoEmotions. We report Accuracy, Macro-F1, and WeightedF1, and extend standard evaluation with robustness tests under realistic perturbations (typos, token dropout, punctuation noise, hashtag transformations). We also perform slice-based diagnostics using heuristic linguistic markers (emoji use, capitalization, pronouns, profanity, negation) to surface performance disparities that aggregate metrics can hide. Results show that fine-tuned BERT performs best on clean data, while prompted LLMs are generally more tolerant to surface noise but have lower absolute accuracy. Slice analysis highlights consistent degradation on profanity and negation, indicating persistent challenges on informal, style-heavy text. These findings provide practical guidance for deploying emotion models in noisy social media settings.
- 14 February 2026
- Published by Springer Nature
- p. 567-587
AI-Augmented Data Science Pipelines for Holistic Business Transformation
- 14 February 2026
- Published by Springer Nature
- p. 567-587
AI-Augmented Data Science Pipelines for Holistic Business Transformation
- 14 November 2025
- Published by IGI Global Scientific Publishing
- p. 201-236
Federated Intelligence for Secure Wireless Systems
This chapter explores federated intelligence as a scalable and privacy-preserving approach to securing wireless systems. With the growing adoption of wireless technologies across critical sectors such as smart cities, transportation, and healthcare, traditional centralized security frameworks are increasingly inadequate. The chapter introduces a modular architecture that combines federated learning, edge AI, and privacy-enhancing techniques like differential privacy and homomorphic encryption. It addresses challenges posed by non-IID data, adversarial behavior, and resource constraints. Real-world use cases and implementation strategies are examined to demonstrate the practical viability of this approach in complex wireless environments.
- 14 November 2025
- Published by IGI Global Scientific Publishing
- p. 201-236
Federated Intelligence for Secure Wireless Systems
This chapter explores federated intelligence as a scalable and privacy-preserving approach to securing wireless systems. With the growing adoption of wireless technologies across critical sectors such as smart cities, transportation, and healthcare, traditional centralized security frameworks are increasingly inadequate. The chapter introduces a modular architecture that combines federated learning, edge AI, and privacy-enhancing techniques like differential privacy and homomorphic encryption. It addresses challenges posed by non-IID data, adversarial behavior, and resource constraints. Real-world use cases and implementation strategies are examined to demonstrate the practical viability of this approach in complex wireless environments.
- Open Access
- 18 July 2025
- Published by Atlantis Press SARL
- p. 563-576
Explainable AI for Sustainability: Bridging Trust, Ethics, and Accountability
- 18 July 2025
- Published by Atlantis Press SARL
- p. 563-576
Open AccessExplainable AI for Sustainability: Bridging Trust, Ethics, and Accountability
- 9 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 93-99
Distributed Data Intelligence in the Edge-to-Cloud Continuum: a Survey of Architectures, Models, and Applications
The exponential growth of data generated across edge devices, sensors, and cloud infrastructures has necessitated a paradigm shift in the distribution, processing, and action of intelligence. Distributed Data Intelligence (DDI) has emerged as a pivotal approach for enabling real-time, privacy-preserving, and scalable analytics across the edge-to-cloud continuum. This survey presents a comprehensive review of DDI systems focusing on their architectural foundations, computational models, and practical applications. It begins by establishing the conceptual underpinnings of the DDI and its evolution from traditional centralized paradigms. A detailed taxonomy of edge-fog-cloud architectures was introduced, highlighting orchestration mechanisms, virtualization technologies, and the role of federated learning. This study further examined lightweight models, hierarchical intelligence, and collaborative AI approaches tailored to distributed environments. Application domains such as smart healthcare, autonomous systems, industrial IoT, and smart cities were explored to illustrate the transformative potential of DDI. Finally, this survey identified critical challenges and open research questions, including interoperability, trust, sustainability, and governance in heterogeneously distributed ecosystems. By synthesizing insights from recent advancements, this study provides a structured foundation and a forward-looking perspective on the future of distributed intelligence in the edge-to-cloud era.
- 9 July 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 93-99
Distributed Data Intelligence in the Edge-to-Cloud Continuum: a Survey of Architectures, Models, and Applications
The exponential growth of data generated across edge devices, sensors, and cloud infrastructures has necessitated a paradigm shift in the distribution, processing, and action of intelligence. Distributed Data Intelligence (DDI) has emerged as a pivotal approach for enabling real-time, privacy-preserving, and scalable analytics across the edge-to-cloud continuum. This survey presents a comprehensive review of DDI systems focusing on their architectural foundations, computational models, and practical applications. It begins by establishing the conceptual underpinnings of the DDI and its evolution from traditional centralized paradigms. A detailed taxonomy of edge-fog-cloud architectures was introduced, highlighting orchestration mechanisms, virtualization technologies, and the role of federated learning. This study further examined lightweight models, hierarchical intelligence, and collaborative AI approaches tailored to distributed environments. Application domains such as smart healthcare, autonomous systems, industrial IoT, and smart cities were explored to illustrate the transformative potential of DDI. Finally, this survey identified critical challenges and open research questions, including interoperability, trust, sustainability, and governance in heterogeneously distributed ecosystems. By synthesizing insights from recent advancements, this study provides a structured foundation and a forward-looking perspective on the future of distributed intelligence in the edge-to-cloud era.
- Open Access
- 2 July 2025
- Published by Penerbit UTM Press in International Journal of Innovation and Business Strategy (IJIBS)
- Vol. 20 (1) , 39-52
The Impact of AI and Machine Learning on Business Model Innovation
Artificial Intelligence (AI) and Machine Learning (ML) are fundamentally reshaping business model innovation across industries. These technologies drive automation, predictive analytics, and hyper-personalized experiences, enabling organizations to optimize operations, create new revenue streams, and enhance customer engagement. This research explores the transformative impact of AI and ML on business models, particularly within platform-based and subscription-based models in industries such as e-commerce and financial services. By analyzing real-world case studies and emerging trends, we highlight how AI Powered innovations, including intelligent automation, the impact of rapidly developing generative AI applications, and data-driven decision-making, foster competitive advantage. Furthermore, AI-driven value capture mechanisms, such as dynamic pricing and adaptive monetization strategies, are redefining revenue generation. While AI adoption presents significant opportunities, challenges such as data privacy concerns, algorithmic bias, and workforce upskilling requirements remain key considerations. This study provides insights into how businesses can effectively integrate AI-driven solutions to achieve sustainable innovation, maintain agility, and navigate the evolving digital economy.
- 2 July 2025
- Published by Penerbit UTM Press in International Journal of Innovation and Business Strategy (IJIBS)
- Vol. 20 (1) , 39-52
Open AccessThe Impact of AI and Machine Learning on Business Model Innovation
Artificial Intelligence (AI) and Machine Learning (ML) are fundamentally reshaping business model innovation across industries. These technologies drive automation, predictive analytics, and hyper-personalized experiences, enabling organizations to optimize operations, create new revenue streams, and enhance customer engagement. This research explores the transformative impact of AI and ML on business models, particularly within platform-based and subscription-based models in industries such as e-commerce and financial services. By analyzing real-world case studies and emerging trends, we highlight how AI Powered innovations, including intelligent automation, the impact of rapidly developing generative AI applications, and data-driven decision-making, foster competitive advantage. Furthermore, AI-driven value capture mechanisms, such as dynamic pricing and adaptive monetization strategies, are redefining revenue generation. While AI adoption presents significant opportunities, challenges such as data privacy concerns, algorithmic bias, and workforce upskilling requirements remain key considerations. This study provides insights into how businesses can effectively integrate AI-driven solutions to achieve sustainable innovation, maintain agility, and navigate the evolving digital economy.
- 1 June 2025
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 72-79
TRANSFORMING ACADEMIC RESEARCH: THE IMPACT OF ARTIFICIAL INTELLIGENCE ON METHODOLOGIES AND COLLABORATION
- 1 June 2025
- Published by IATED Academy in EDULEARN Proceedings
- Vol. 1, 72-79
TRANSFORMING ACADEMIC RESEARCH: THE IMPACT OF ARTIFICIAL INTELLIGENCE ON METHODOLOGIES AND COLLABORATION
- Open Access
- 20 May 2025
- Published by Institute of Central Computation and Knowledge Inc. in ICCK Transactions on Machine Intelligence
- Vol. 1 (1) , 6-16
Privacy-Preserving Federated Learning for IoT Botnet Detection: A Federated Averaging Approach
Traditional centralized machine learning approaches for IoT botnet detection pose significant privacy risks, as they require transmitting sensitive device data to a central server. This study presents a privacy-preserving Federated Learning (FL) approach that employs Federated Averaging (FedAvg) to detect prevalent botnet attacks, such as Mirai and Gafgyt, while ensuring that raw data remain on local IoT devices. Using the N-BaIoT dataset, which contains real-world benign and malicious traffic, we evaluated both the IID and non-IID data distributions to assess the effects of decentralized training. Our approach achieved 97.1% F1-score in IID and 94.8% in highly skewed non-IID scenarios, closely matching centralized learning performance while preserving privacy. Additionally, communication optimization techniques—Top-20% gradient sparsification and 8-bit quantization—reduce communication overhead by up to 80%, significantly enhancing the efficiency. Our convergence analysis further shows that FedAvg remains effective under non-IID conditions, thereby demonstrating its robustness for real-world deployments. These results demonstrate that FL provides a scalable and privacy-preserving solution for securing IoT networks against botnet threats.
- 20 May 2025
- Published by Institute of Central Computation and Knowledge Inc. in ICCK Transactions on Machine Intelligence
- Vol. 1 (1) , 6-16
Open AccessPrivacy-Preserving Federated Learning for IoT Botnet Detection: A Federated Averaging Approach
Traditional centralized machine learning approaches for IoT botnet detection pose significant privacy risks, as they require transmitting sensitive device data to a central server. This study presents a privacy-preserving Federated Learning (FL) approach that employs Federated Averaging (FedAvg) to detect prevalent botnet attacks, such as Mirai and Gafgyt, while ensuring that raw data remain on local IoT devices. Using the N-BaIoT dataset, which contains real-world benign and malicious traffic, we evaluated both the IID and non-IID data distributions to assess the effects of decentralized training. Our approach achieved 97.1% F1-score in IID and 94.8% in highly skewed non-IID scenarios, closely matching centralized learning performance while preserving privacy. Additionally, communication optimization techniques—Top-20% gradient sparsification and 8-bit quantization—reduce communication overhead by up to 80%, significantly enhancing the efficiency. Our convergence analysis further shows that FedAvg remains effective under non-IID conditions, thereby demonstrating its robustness for real-world deployments. These results demonstrate that FL provides a scalable and privacy-preserving solution for securing IoT networks against botnet threats.
- Open Access
- 8 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1696-1704
Robust Defense Mechanisms for Agentic News Recommenders: Mitigating Data Poisoning Attacks
- 8 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1696-1704
Open AccessRobust Defense Mechanisms for Agentic News Recommenders: Mitigating Data Poisoning Attacks
- Open Access
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Fault-Tolerant Federated Learning Framework for Edge Devices in Unstable Networks
Federated Learning (FL) has emerged as a critical paradigm for decentralized machine learning, enabling collaborative model training across edge devices, while preserving data privacy. However, the real-world deployment of edge devices faces challenges such as intermittent connectivity, resource heterogeneity, frequent client dropouts, and Byzantine failures, which severely affect the reliability and convergence of FL systems. To address these challenges, we propose REFINE (Robust and Efficient Federated learning for unstable networked edge devices), which is a fault-tolerant FL framework designed specifically for unreliable edge environments. REFINE integrates asynchronous communication, availability-aware client selection, hierarchical aggregation via Edge Aggregators (EAs), and adaptive trimmed mean filtering to mitigate malicious updates. It leverages lightweight client scoring and robust statistical techniques to prioritize reliable contributions while filtering for outliers. Extensive evaluations on CIFAR-10, FashionMNIST, and Human Activity Recognition datasets under various fault conditions, including client dropout rates up to 50%, network instability, and Byzantine attacks-demonstrate that REFINE achieves up to 11.4% higher accuracy and 30% faster convergence compared to baseline methods such as FedAvg, FedAsync, and Fed-MS, while maintaining low overhead. REFINE offers a practical and scalable solution that enables resilient and trustworthy federated learning across unstable edge networks.
- 1 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in TechRxiv
Open AccessFault-Tolerant Federated Learning Framework for Edge Devices in Unstable Networks
Federated Learning (FL) has emerged as a critical paradigm for decentralized machine learning, enabling collaborative model training across edge devices, while preserving data privacy. However, the real-world deployment of edge devices faces challenges such as intermittent connectivity, resource heterogeneity, frequent client dropouts, and Byzantine failures, which severely affect the reliability and convergence of FL systems. To address these challenges, we propose REFINE (Robust and Efficient Federated learning for unstable networked edge devices), which is a fault-tolerant FL framework designed specifically for unreliable edge environments. REFINE integrates asynchronous communication, availability-aware client selection, hierarchical aggregation via Edge Aggregators (EAs), and adaptive trimmed mean filtering to mitigate malicious updates. It leverages lightweight client scoring and robust statistical techniques to prioritize reliable contributions while filtering for outliers. Extensive evaluations on CIFAR-10, FashionMNIST, and Human Activity Recognition datasets under various fault conditions, including client dropout rates up to 50%, network instability, and Byzantine attacks-demonstrate that REFINE achieves up to 11.4% higher accuracy and 30% faster convergence compared to baseline methods such as FedAvg, FedAsync, and Fed-MS, while maintaining low overhead. REFINE offers a practical and scalable solution that enables resilient and trustworthy federated learning across unstable edge networks.