Sarojini Naidu College for Women

This article discusses a topology adaptive self-organizing circular neural network (TASOCNN) that can be used to cluster circular data. TASOCNN consists of a set of interconnected processors that are trained to capture the topological structure of the circular data. The connections (edges) between the processors are assigned strength values which are updated during the training process. However, the naive TASOCNN model has some limitations, such as its inability to accurately distinguish between inter-cluster and intra-cluster edges. The article proposes significant modifications to naive TASOCNN, including updates to strength equations for generating stronger and shorter edges in dense areas of data. Additionally, a novel non-iterative clustering procedure based on a two-component Beta mixture model is introduced. This model isolates intra-cluster edges by analyzing strength values, leading to the removal of inter-cluster edges. The application of TASOCNN is demonstrated in the context of color image segmentation, a crucial task in various applications. The results show that the proposed method outperforms the naive TASOCNN model and other state-of-the-art self-organizing neural models as well as mixture models.

The global context of education has brought a new dimension to development education to build an inclusive sustainable future. Educational Interventions in the form of developmental education builds a sense of consciousness, skills of analysis and understanding, along with efficiency to promote sustainability and justice. Sustainability includes a wide range of parameters including environmental factors, social and cultural factors, which can only be attained by creating awareness and empower the learners to take responsibility to bring a transition, brought through this sustainable, education system. This can be attained through development education which is needed to address the inequalities and social injustice embedded in our society. This paper is based on a model which shows that the three factors ‘people– planet - profit’ (socio – economic - environmental) framework which is being adopted to study the complexities of different aspects of Primary Education system in West Bengal. The sustainable education based on this framework would imply the three basic factors i.e. sustainable educational policies, enhance the competency of teachers, learners and other stakeholders in the society and enrich the socio-environmental eco-system of educational institutions to attain the desirable sustainable outcome. This framework is being applied to see whether surveyed data from different primary schools of different villages in North 24 Parganas of West Bengal, would lead to sustainable education. This paper incorporates the three factors from the model and the corresponding transformation that it can bring about through these factors of education to attain sustainability in primary education system. Results show the Issues of Driver 1 represent, Information on Quality Indicators which is represented by the availability of learning resources, Issues of Driver 2 represent the community participation from various categories of students, represented by the engagement of various categories of students and teacher’s whose pro-active role is enhanced through training and skills along with community participation. The Issues of Driver 3 (Planet) constitute creating a favourable environment for sustainable education represented by creating an environment-friendly atmosphere leading to sustainable education in the primary schools. All these three drivers work harmoniously to create a sustainable education system.

Artificial Intelligence (AI) and Quantum Computing (QC) represent the most significant transformative technological advancements of the 21st century. While AI excels in optimization and decision-making, QC promises supreme computational potential via superposition and entanglement. The confluence of these fields known as AI-Quantum Integration, is emerging as a critical frontier for solving complex, data-intensive problems across domains such as drug discovery and financial analytics. This paper proposes a novel computational framework, the Hybrid AI-Quantum Architecture (HAIQF), which combines quantum variational circuits (VQC) fine-tuned to classical learning algorithms. This hybrid approach is designed to mitigate critical challenges currently facing both fields including computational efficiency, low gate fidelity and noise limitations inherent in Noisy Intermediate-Scale Quantum (NISQ) devices. The study provides a comprehensive review of the theoretical and empirical landscape of Quantum Machine Learning (QML), detailing methodologies like VQC, Quantum Kernel Methods and Quantum Neural Networks. The core of the paper focuses on analyzing the effectiveness of the HAIQF model through rigorous experimental results and benchmarks. Ultimately, the goal is to conduct a comparative analysis of the HAIQF model's performance against the broader ethical, technical, and real-world implications of integrating these two powerful modalities.

Clustering algorithms play an important role in movie recommendation systems by allowing the grouping of similar users or movies together. This grouping facilitates more accurate and personalized recommendations. Clustering contributes to a better user experience by delivering more relevant and timely recommendations. Users are more likely to engage with recommendations that closely match their interests and needs. K-means clustering is a powerful algorithm in this context, facilitating personalized recommendations. This paper presents a novel movie recommendation system, Demographic-KM, which utilizes a customized K-means algorithm that leverages user demographic information. The proposed approach uses K-means clustering to segment users and movies into clusters and makes recommendations based on these clusters. It creates a user clustering system based on demographic information and a movie clustering system based on the genres. It maps user clusters to movie clusters using rating data and recommends movies to new users based on their predicted clusters. This association allows the recommendation system to suggest movies from the most suitable movie cluster for a new user, determined by their predicted user cluster. The clustering quality is evaluated using silhouette scores, root mean squared error (RMSE), and mean absolute error (MAE). The proposed method shows low MAE and RMSE values and high silhouette scores, indicating strong clustering performance as well as improved recommendation.

Sensitive quantitative characteristics, such as household income, incidence of premarital abortion or substance use, are difficult to reliably measure, because respondents frequently distort or withhold their true value for fear of stigma. Scrambled response (SR) techniques address the issue by allowing a respondent to mask an answer by a randomization device before it is reported. However, current SR estimators for heterogeneous populations have been mostly developed either under simple random sampling or under unstratified ranked set sampling and have generally used a single additive or multiplicative scrambling mechanism rather than a combined one. In this paper, we develop a new scrambled response strategy under stratified ranked set sampling (SRSS). The sensitive response Y is scrambled through a combined multiplicative-additive transformation Z = UY + V, which includes the classical additive [21] and multiplicative [11] scrambling schemes as special cases. The scrambled output data are used to propose a class of almost unbiased ratio type estimators of the population mean indexed by three scalars chosen so that the weights sum to one, the first order bias vanishes and the mean square error is minimized; the bias and mean square error of the resulting estimator are derived to the first order of approximation and the optimum member of the class is obtained in closed form. The efficiency of the proposed estimator is assessed in terms of percent relative efficiency and percent relative loss with respect to the corresponding non-scrambled stratified ranked set sampling estimator based on a real population of U.S. state-level abortion rates and an artificially generated population under several configurations of the scrambling-variable parameters. In all the configurations considered, the proposed estimator is more efficient than the non-scrambled one, which means that the loss of precision incurred by scrambling to protect the respondent’s privacy is more than compensated by the joint use of stratification, ranking and the flexible three-parameter estimator class. The results provide a basis for recommending the proposed strategy for surveying practitioners who deal with sensitive quantitative data from heterogeneous populations. The paper ends with a discussion of extensions to imperfect ranking and higher-order approximations as future work.