Sowmya T

Dr. Sowmya T


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 April 2025 in MethodsX
Verified
Journal Article
Open access
RDRaghavendra M Devadas
Raghavendra M Devadas
STSowmya T
Sowmya T
Quantum Machine Learning (QML) is the emerging confluence of quantum computing and artificial intelligence that promises to solve computational problems inaccessible to classical systems. Using quantum principles such as superposition, entanglement, and interference, QML promises exponential speed-ups and new paradigms for data processing in machine learning tasks. This review gives an overview of QML, from advancements in quantum-enhanced classical ML to native quantum algorithms and hybrid quantum-classical frameworks. It varies from applications in optimization, drug discovery, and quantum-secured communications, showcasing how QML can change healthcare, finance, and logistics industries. Even though this approach holds so much promise, significant challenges remain to be addressed-noisy qubits, error correction, and limitations in data encoding-that must be overcome by interdisciplinary research soon. The paper tries to collate the state of the art of QML in theoretical underpinnings, practical applications, and directions into the future.

25 March 2026 in Energy Informatics
Verified
Journal Article
Open access
PDF file
RDRaghavendra M. Devadas
Raghavendra M. Devadas
STSowmya T
Sowmya T
The increasing computational requirements in modern Artificial Intelligence (AI) development have raised the stakes when it comes to the environmental sustainability of machine learning use and application. The topic that has received relatively little treatment in previous research is the carbon footprint of AI systems, but it is not studied continuously across algorithms, hardware, life cycle, and use. This work takes stock of the state-of-the-art to provide a comprehensive review of carbon-aware AI from the bottom up for the entire gamut of computation. Consistent with PRISMA principles, by way of systematic search across prominent academic databases and repositories (2018–2025), we identified 784 unique citations and extracted 62 studies that satisfied pre-established inclusion criteria. These studies were organized by four domains: algorithms for energy (20), hardware and accelerators (15), Life-cycle assessment (LCA) (9), and operation under deployment (11). Three major conclusions follow from the synthesis. First, algorithmic efficiency – including pruning, quantization, and sparsity might reduce computational burden to meet carbon goals; however, they only achieve concrete carbon reduction when accounted for in hardware and data centre setups. Second, life-cycle analyses show that while operational energy continues to account for most emissions during large-scale training, the embodied carbon from semiconductor fabrication plays an increasingly important role in fleets with a lot of equipment or frequent refreshes. Third, deployment decisions such as data center location, carbon-aware scheduling, and cloud–edge workloads placement bring much more variance on real emissions compared to what can be achieved at model-level optimisation. Between sectors, inconsistencies in methodology – notably for carbon reporting, system boundaries, and energy telemetry – hinder reproducibility and comparison of findings across studies. To overcome these limitations, this review suggests a research agenda with respect to standardized carbon accounting, hardware–software co-optimization, better available embodied-emission data, and the inclusion of carbon in decision-making for AutoML and scheduling systems. This comprehensive integration will serve as a stepping-stone towards the development of sustainable AI across academia and industry.

7 July 2025 in IEEE Access
Verified
Journal Article
Open access
STSowmya T
Sowmya T
EAE. A. Mary Anita
E. A. Mary Anita
In this paper, network intrusion detection is proposed using an improved version of the support vector machine model to detect DoS attacks. Here, the SVM model considers the weight parameter along with the kernel to find the best decision boundary that separates the data into DoS and normal. The proposed model provides a novel kernel trick that reduces the overlapping of data. The intrusion detection system aims to construct an ideal system that can detect attacks with very high performance using a ShiP vector machine(Sophisticated High Performance Vector Machine). The framework comprises three major steps: data collection and preprocessing, Recursive Feature Elimination (RFE) based feature selection, and the ShiP Vector Machine classification strategy. The system is evaluated using the DoS dataset from UNSWNB15 and real time PSD-23 sniffer dataset. DoS data is generated by extracting the normal and DoS attacks from the UNSWNB15 dataset. Experimental results show that the proposed ShiP vector machine shows outstanding performance by achieving 96.44 % accuracy on the DoS dataset and 90.12 % accuracy for real time PSD-23 data.

12 June 2026
Verified
Book Chapter
Open access
SBSomnath Bhattacharya
Somnath Bhattacharya
PGPraveen Gujjar
Praveen Gujjar
RDRaghavendra Devadas
Raghavendra Devadas
VHVani Hiremani
Vani Hiremani
STSowmya T
Sowmya T
RSR. Sapna
R. Sapna
Deep ensemble learning has become a very popular and efficient technique for making predictions. Traditional machine learning models are often not efficient when used for making predictions of imbalanced datasets. To counter the effects of imbalanced datasets, several different oversampling and synthetic sampling methods have been proposed. The effect of such sampling methods on deep ensemble learning methods is also not well studied. This study aims to explore two basic questions: whether deep ensemble models are suitable for unbalanced data given traditional methods are not and secondly if we use the oversampling method along with deep ensemble learning, does it improve model performance? For our comparison, we choose four models. We further scope this study by using only sequential ensemble models. We choose three baseline models and two deep-learning ensemble models for our study. We find that the deep ensemble learning models are better when we consider a combination of precision and recall. Also, oversampling significantly increases the model’s ability to correctly predict the underrepresented class.


18 August 2026 in Journal of King Saud University Computer and Information Sciences
Verified
Journal Article
Open access
PDF file
RDRaghavendra Devadas
Raghavendra Devadas
STSowmya T
Sowmya T
Quantum technologies — encompassing quantum sensing, quantum computing and machine learning (QML), quantum nanomaterials, and quantum cryptography — represent an emerging class of physical and computational tools with the potential to address fundamental limitations in clinical medicine across diagnostics, therapeutics, drug discovery, and health-information security. Despite a rapidly growing body of peer-reviewed evidence, estimated at a compound annual growth rate of approximately 14.8% over 2016–2025, the extant literature organises findings by technology type rather than clinical specialty, limiting accessibility for clinically-oriented readers and precluding systematic cross-domain technology maturity comparison. This review addresses that structural gap through a specialty-first synthesis of the quantum-medicine literature and the introduction of an original five-dimension Translational Readiness Taxonomy (TRT: Hardware Maturity, Validation Setting, Benchmarking Rigour, Reproducibility, Regulatory Pathway) as a replicable cross-domain maturity assessment instrument. A total of 168 peer-reviewed studies were identified through a reproducible seven-step PRISMA-aligned screening pipeline from a deduplicated corpus of 14,303 records (Scopus n = 7,081; Web of Science n = 4,644; PubMed n = 2,578), organised across five clinical domains: cardiology , neurology , oncology, genomics and drug discovery , and health data security ; 2022; Indumathi et al. 2024; Roosan et al. 2025; Freyer et al. 2025; Roosan et al. 2024; Lo et al. 2024; Singh et al. . Three original synthesising frameworks are additionally introduced: the Quantum-Medicine Translation Pipeline (QMTP), the Evidence-Maturity Bubble Matrix, and the Quantum-Medicine Ecosystem Map. Neurology leads translational readiness across hardware maturity, clinical validation, and benchmarking dimensions, driven by progressive optically-pumped-magnetometer magnetoencephalography (OPM-MEG) clinical adoption. Oncology, despite constituting 41.7% of included studies and generating the four highest-cited records, exhibits the lowest TRT composite score, with no prospective patient-recruited trials identified across its 70 included studies. A QML-framing-versus-evidence gap is confirmed in cardiology and oncology, where near-term clinical QML claims lack the computational proof-of-concept infrastructure that would logically need to precede them. Health data security leads regulatory maturity via the 2024 NIST finalisation of post-quantum cryptographic standards (CRYSTALS-Kyber/FIPS 203)