Bhagyashree Hambarde’s research while affiliated with St Vincent Pallotti College of Engineering & Technology and other places

Generative Adversarial Networks (GANs), introduced by Ian Goodfellow in 2014, have revolutionized machine learning by creating new data. GANs consist of two parts: the Generator, which creates data samples from random noise, and the Discriminator, which distinguishes real from fake data. These networks compete, with the Generator improving its fakes and the Discriminator enhancing its detection. This competition drives both networks to improve. Training GANs can be challenging due to issues like instability and mode collapse. Advanced versions like Conditional GANs (c GANs) and Wasserstein GANs (WGANs) address these problems. This chapter introduces GANs, explaining their components, adversarial training, and practical implementation using the MNIST dataset. It includes guidance on tools, dataset understanding, architecture design, and visualization techniques for evaluating generated images.

The rapid advancement of financial technology (FinTech) has led to the integration of advanced technologies like data science, blockchain, cloud computing, and artificial intelligence. However, trust evaluation remains a critical challenge in dynamic landscape. Existing trust evaluation methods often neglect key aspects of timeliness, reliability, and non‐invasiveness, leading to imprecise trust assessments and insufficient detection of malicious user behavior. This paper introduces a robust four‐layer architectural framework with the blockchain layer, edge computing service layer, cloud computing service layer, and terminal user application layer leveraging blockchain technology for authentication and trust evaluation. Blockchain technology transforms FinTech data into linked data, ensuring data security and decentralization during information transfers. A novel hybrid consensus protocol combining Proof of Elapsed Time (PoET) and Proof of Stake (PoS) is introduced to enhance the efficiency and security of the blockchain. Extensive simulation experiments have demonstrated significant improvements in data security, reliability, and accuracy of trust assessments compared to existing methods. This paper presents a comprehensive solution for enhancing trust evaluation in FinTech, emphasizing timeliness, reliability, and non‐invasiveness of assessments.

Protecting the environment and ensuring the availability of potable water requires efficient wastewater treatment. This paper investigates the need for an advanced process optimization model to enhance the efficacy of wastewater treatment processes. Existing optimization models for wastewater treatment are frequently incapable of effectively analyzing historical data, optimizing dosages, or predicting optimal process parameters. To circumvent these restrictions, a novel method for optimizing various aspects of the treatment procedure using auto encoders (AE), genetic algorithm (GA), and vector autoregressive moving average (VARMA) models is proposed. Extensive experimentation with multiple datasets and samples, including the Melbourne Wastewater Treatment Dataset, Urban Wastewater Treatment, and Global Wastewater Treatment, demonstrates significant improvements in our proposed model. Compared to recently proposed models, our method results in an average improvement of 8.5% in treatment quality, a 4.9% reduction in delay, and a 9.5% increase in water purity. Beyond the datasets mentioned, our model's applications and use cases provide a valuable framework for optimizing wastewater treatment in a variety of settings. The adaptability and effectiveness of the proposed model make it suitable for both small and large treatment plants.

The optic nerve damaging condition called Glaucoma. This disease is increment at an alarming rate. By the end of the 2044 there is possibility that across 111.8 million populations will be influenced by glaucoma. It is a neurodegenerative disease. If intravascular pressure is increases, optic nerve of the eye gets damage. This damage may cause permanent or total blindness in person. The Glaucoma is examined by an experienced ophthalmologist on the retinal part of the eye. This process required excessive equipment, experienced medical practitioners and also it take more time to work out manually. After considering this problem there is an extreme requirement of developing an automatic system which will effectively and automatically work properly in lack of any professional doctor and it should also take less time. Lots of different parameters are available to detect glaucoma but thebest parameter is to find out optical cup-to-disc-ratio. To increase or to enhance the precision and accuracy of the result, cup to disc value is needed to find CDR value. In order to detect glaucoma, automatic separation of the OC and DC is very essential to avoid any error. We use deeplabv3 architecture to perform segmentation of optic disc and cup and classification is done using ensemble machine learning. This proposes research achieve intersection over union (IOU) scores, 0.9423 for optic disc and 0.9310 for optic cup. We perform testing on globally accessible data-sets i.e. DRISHTI, ORIGA, and RIMONE with accuracy of 93%, 91% and 92% respectively