Confluence: A Journal of Technology, Business and Society

A TECHNICAL COMPARISON OF GANS, VARIATIONAL AUTOENCODERS, AND DIFFUSION MODELS IN MULTIMODAL CONTENT GENERATION

Basu Dev Shivhare, Radha Raman Chandan — Fri, 07 Aug 2026 00:00:00 +0000

In this study, we compare and contrast Generative Adversarial Networks, Variational Autoencoders and Diffusion Models in three central evaluation metrics: sample fidelity, controllability and training stability. It will cover the adversarial minimax formulation that has been the foundation for Generative Adversarial Networks, the training instabilities such as mode collapse which have historically hindered their reliability and the architectural and loss function modifications that have been made to improve the issues associated with such instabilities.

Based on the analysis, Diffusion Models currently have the best trade-off between sample fidelity and conditional controllability, with high inference latency but low training instability, Variational Autoencoders have the best sample fidelity and a stable training process with an interpretable latent space but lower inference speed, and Generative Adversarial Networks have the lowest training instability and inference speed but a less stable training process due to persistent training instability issues.

Fiscal Consolidation Vs Welfare Expansion in India’s Budget Framework

Anju Singh, Jyoti Bery — Fri, 07 Aug 2026 11:56:11 +0000

The budgetary system in India is a constant policy dilemma between fiscal tightening and welfare growth. On the one hand, fiscal prudence is expected to decrease the deficit level, stabilize the government debt, and decrease inflationary pressures, as well as promote the credibility of the macro economy. In contrast, welfare-based spending, i.e. specific subsidies, income support programs, rural development, health insurance, food security, etc. are still the core of inclusive growth and poverty elimination. This paper focuses on India and the way in which the nation achieves a balance between the goals of reducing deficit under the standards of fiscal responsibility and the increased necessity to spend on the social sector in a developing economy. The paper analyses the trend of revenue deficits, fiscal deficits, capital expenditure, and social spending trends using secondary data based on the Union Budget documents, fiscal indicators, and the major welfare allocations during the last ten years. It discusses the merger of strategies as crowding out welfare commitments or structural reforms, tax buoyancy and rationalization of subsidies facilitate all three to pursue fiscal discipline and social investment at the same time. The results provide an indication that the reconciliation of fiscal sustainability and social protection goals is possible through a calibrated strategy, which is to focus on productive capital expenditure and to rationalize the inefficient subsidies. The paper concludes that growth in the long term, institutional credibility and welfare design targeted are the key in the realization of the balanced and robust budget structure in India.

Machine Learning-Driven Predictive Maintenance Models for Industrial Equipment and Manufacturing Systems

Amar Deep Gupta, Basu Dev Shivhare — Fri, 07 Aug 2026 12:04:13 +0000

There has been an escalation in the complexity of industrial machines and factory structures which have exacerbated the necessity of smart maintenance systems that can minimize unexpected downtime and operation expenses. This study is a detailed work on the predictive maintenance models with machine learning, applying to a multivariate sensor data of industrial machinery. Data-driven models of fault classification, anomaly detection and the prediction of the remaining useful life have been developed using vibration, temperature, pressure, acoustic and electrical signals. Four machine learning algorithms, namely Random Forest, Support Vector machine, Long Short-term memory (LSTM), and Isolation Forest were applied and tested in the context of the realistic industrial setting. As shown in the experimental results LSTM model has provided the best accuracy in classification of fault of 95.1 percent with an F1-score of 0.94 which is best in the sense that it is able to capture the temporal degradation patterns. Random Forest model also performed well with an accuracy of 94.2 which is better in providing better interpretability and fast inference. In unsupervised anomaly detection, Isolation Forest reported the highest rate of detection as 88.7% and false alarm low rate of 6.4% and therefore is feasible in early fault detection when there are scarce labeled data. In the case of the useful life prediction, the LSTM model minimized the error involved in prediction, whereby its mean absolute error was 11.2 hours, which is lower than 18.6 hours in case of the conventional models. On the whole, the findings prove the idea that machine learning-based predictive maintenance contributes to a stronger degree of reliability, optimization of the maintenance process, and helps to make decisions based on the available data in the context of the contemporary manufacturing system.

Comparative Analysis of Feature Selection and Dimensionality Reduction Techniques for High-Dimensional Datasets

Sushil Kumar — Sat, 08 Aug 2026 06:08:09 +0000

The existence of high-dimensional data is becoming widespread in fields including healthcare, bioinformatics, text analytics, and factory diagnostics, and has created difficulties in computational complexity, data redundancy, and model overfitting. Learning efficiency and predictive performance hence require effective techniques of feature selection and dimensionality reduction in order to enhance the efficiency of learning. The paper includes a comparative analysis of four popular methods, such as Chi-Square feature selection, Recursive Feature Elimination (RFE), Principal Component Analysis (PCA), and Linear Discriminant Analysis (LDA), on several high dimensions of data (gene expression, text and medical imaging data) and assessed their effectiveness. There are experimental findings that show that supervised techniques are always better as compared to unsupervised and filter techniques. The maximum classification accuracy of LDA was 90.5, precision was 89.7, recall was 89.2 and F1-score was 89.5 which minimized the feature space by 99.98. There was also good performance of RFE with an average accuracy of 88.9 but at a greater cost of computation. On the contrary, Chi-Square selection was the quickest to execute (in terms of time) (about 1.0 s) but less accurate (84.6%). PCA efficiently and accurately (balance between efficiency and performance) retained 95 percent variance. Altogether, the results show that the effectiveness of dimensionality reduction methods strongly depends on the data characteristics and the goal of application that can provide a reasonable guide to the choice of the adequate methods when working with big data set.

TruthLens: Graph Analytics for Fake News Detection

Ayushi Shukla, Kushagra Kumar Porwal, Basu Dev Shivhare — Sat, 08 Aug 2026 06:19:36 +0000

Fake news has evolved from a minor online problem to a potent instrument for influencing public opinion and undermining confidence in digital platforms. Conventional detection systems function similarly to linguistic spell checkers, but they frequently overlook the intricate, well-planned nature of disinformation campaigns. In order to close this gap, TruthLens proposes a hybrid, multimodal framework that integrates graph analytics, computer vision, and natural language processing (NLP). Our approach makes use of PyTorch Geometric (GNNs) to monitor the propagation of information in social networks, BERT for comprehensive semantic text analysis, and CLIP for consistency between text and images. TruthLens identifies structural anomalies that indicate bot networks and organized influence operations by examining not only what is said but also how and by whom it spreads. According to preliminary tests, this layered approach performs significantly better than conventional text-only models, providing both greater accuracy and lucid insights into the mechanisms of digital deception.