Ujjwal

Ujjwal

Auto-generated profile, based on publicly available information

Ujjwal

Manipal Institute of Technology, Manipal Academy of Higher Education, Manipal 576104, India

Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

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

Representative Publications

View All Publications
13 October 2023 in IEEE Access
Verified
Journal Article
Open access
UUUjjwal Ujjwal
Ujjwal Ujjwal
RKRajkishor Kumar
Rajkishor Kumar
RBRadhakrishna Bhat
Radhakrishna Bhat
In this article, we proposed two schemes for generating a broadband optical frequency comb (OFC) source using dual laser sources. In the first proposed technique, the OFC is generated using a cascaded single-drive Mach-Zehnder modulator (MZM) and dual-drive Mach-Zehnder modulator (DD-MZM). Here, the two CW light sources are modulated by each of the MZM, respectively. The first single-drive MZM generates the subcarriers, while The second DD-MZM carries out intensity modulation on the output spectrum generated by the MZM, ensuring that the power fluctuations of both even and odd sidebands are adjusted to an equal level. The second approach involves replacing the single-drive MZM with a phase modulator (PM). The center frequencies of laser 1 and laser 2 must be carefully set to achieve a non-overlapping or gap-free comb spectrum. The frequency of the RF signal driving the modulators determines the spacing between the laser frequencies. In order to illustrate this aspect, we have utilized three distinct RF signal frequencies: 10 GHz, 16 GHz, and 20 GHz. Furthermore, we have examined the constraints that impact the estimation of the center frequency of laser 2, aiming for a non-overlapping and gap-free comb spectrum. By considering these constraints, we can ensure the successful generation of a uniform and stable comb spectrum with the desired characteristics in our proposed approach. The theoretical and primary analysis of cascaded single-drive MZM and DD-MZM, as well as cascaded PM and DD-MZM, is also investigated, and simulation results confirm the findings. By employing our proposed approach, OFC achieves a significant number of comb lines while exhibiting a power deviation of ~ 2 dB in comparison to single laser configurations.

22 August 2025 in IEEE Access
Verified
Journal Article
Open access
ASAparna Saju
Aparna Saju
SGSHIVAPRASAD GUNDIBAIL
SHIVAPRASAD GUNDIBAIL
UUUjjwal Ujjwal
Ujjwal Ujjwal
Mental health disorders are becoming a common concern among adults and children alike.With the growth of cloud computing and the rising concerns over data privacy, integrating machine learning (ML) with secure cloud infrastructure for mental health prediction helps in early detection and timely intervention. In this paper, we propose a novel approach of integrating a voting-based ensemble learning with AES-256-encrypted cloud infrastructure on AWS for a secure and scalable mental health prediction. The ensemble model combines five popular machine learning classifiers: Logistic Regression, Naïve Bayes, SupportVector Machine (SVM), XGBoost, and Random Forest. The proposed model was constructed using these classifiers as its base learners. The system was trained on a dataset of 7,731 Reddit posts related to mental health. We have employed AWS S3 storage and AES-256 encryption for data security. To evaluate the effectiveness of the proposed model, we compared its performance against individual classifiers using a range of evaluation metrics, including accuracy, precision, recall, and F1-score. The results demonstrate that the ensemble model consistently outperforms the individual classifiers, achieving superior performance across all metrics. While Reddit posts may not reflect the language or tone used on other platforms—such as clinical notes, journals, or conversation transcripts—thereby limiting the overall generalizability of the model, it demonstrates high accuracy, scalability, and security, making it a viable choice for deployment in real-world clinical settings for early mental health screening.

4 April 2026
Verified
Thesis
Closed access
UUUjjwal Ujjwal
Ujjwal Ujjwal
Gestion du compromis vitesse-précision dans les systèmes de détection de piétons basés sur apprentissage profond: L'objectif principal de cette thèse est d'améliorer la précision des systèmes de détection de piétons à partir d'image, basés sur l'apprentissage profond sans sacrifier à la vitesse de détection. Pour ce faire, nous effectuons d'abord une analyse quantitative systématique des diverses techniques de détection de piétons à partir d'image. Cette analyse nous permet d'identifier les configurations optimales des différentes composantes d'un système de détection de piétons. Nous examinons ensuite la question de la sélection des meilleures couches convolutionnelles pour extraire les caractéristiques visuelles pour la détection des piétons et proposons un système appelé Multiple-RPN, qui combine plusieurs couches convolutives simultanément. Nous proposons le système Multiple-RPN en deux configurations - une fusion tôt et une fusion-tardive ; nous démontrons ensuite que la fusion-tôt est la plus performante, en particulier pour la détection de piétons de petites tailles et les cas d'occultation de piétons. Cette étude fournit aussi une évaluation quantitative de la sélection des couches convolutionnelles. Nous intégrons ensuite l'approche de la fusion-tôt avec une étape de segmentation pseudo-sémantique pour réduire le cout de traitement. Dans cette approche, la segmentation pseudo-sémantique permet de réduire les faux positifs et les faux négatifs. Ceci, associé à un nombre réduit d'opérations, permet d'améliorer simultanément les performances de détection et la vitesse de traitement (~20 images/seconde) ; les performances sont compétitives avec celles de l'état de l'art sur les bases de données caltech-raisonable (3,79% de taux d'erreurs) et citypersons (7,19% de taux d'erreurs). La dernière contribution de cette thèse est la proposition d'une couche de classification des détections potentielles, qui réduit encore le nombre d'opérations de détection. Il en résulte une réduction de la vitesse de détection (~40 images/seconde) avec une perte minime de performance de détection (3,99% et 8,12% de taux d'erreurs dans les bases de données caltech-raisonable et citypersons respectivement) ce qui reste compétitif avec l'état de l'art.

22 September 2026 in Healthcare
Verified
Journal Article
Open access
PDF file
GHGaurav Chandrashekhar Hajare
Gaurav Chandrashekhar Hajare
VSVenugopal Shanbhogue
Venugopal Shanbhogue
UUUjjwal Ujjwal
Ujjwal Ujjwal
VSVishwajeet Singh
Vishwajeet Singh
Objectives: This study assesses district-level disparities in maternal and child health performance across nine districts of Manipur using the CRITIC–TOPSIS approach. It aims to provide a multidimensional assessment of relative district performance and identify patterns of inter-district disparities to support differentiated maternal and child health planning. Methods: Eleven indicators relating to antenatal care, institutional delivery, postnatal care, immunization, breastfeeding, and nutritional outcomes of children were analyzed using data from the National Family Health Survey-5 (NFHS-5), 2019–2021. The CRITIC method was used to obtain objective weights by considering indicator variability and intercriteria correlations, and TOPSIS was used to construct the composite index and rank the districts according to their relative proximity to an ideal health performance profile. K-means clustering of the standardized composite index was used to explore relative district groupings, with alternative cluster solutions, Ward.D2 hierarchical clustering, and random-initialization stability used to assess the clustering results. Results: The results revealed substantial inter-district variation, with Imphal West having the highest composite index (0.8674), followed by Bishnupur (0.5851) and Imphal East (0.5564), while the lowest value was observed for Ukhrul (0.3499). Strong positive associations were observed among several components of the maternal and newborn continuum of care, particularly institutional delivery and maternal and newborn postnatal care. K-means clustering produced a three-cluster grouping comprising Imphal West in the Higher-Performance cluster; Bishnupur, Imphal East, Churachandpur, Thoubal, and Tamenglong in the Intermediate-Performance cluster; and Senapati, Chandel, and Ukhrul in the Lower-Performance cluster. Although K = 2 yielded the highest silhouette width, it primarily separated Imphal West from all remaining districts. Therefore, K = 3 was retained as an exploratory grouping because it provided greater differentiation among the remaining districts while maintaining positive silhouette values and showing complete agreement with Ward.D2 hierarchical clustering and stability across random initializations. Conclusions: The findings reveal a more nuanced pattern of maternal and child health performance than a simple geographical dichotomy and highlight the value of district-specific assessment. The CRITIC–TOPSIS framework, complemented by cluster analysis, provides a reproducible approach for identifying relative disparities and supporting differentiated maternal and child health planning.