http://localhost:8081/jspui/handle/123456789/15626 2026-08-14T23:58:50Z http://localhost:8081/jspui/handle/123456789/21313 Title: Online Learning for Noisy Polynomial Authors: Chauhan, Deepanshu Abstract: We study deep contextual bandits, a class of contextual bandits where each context-arm pair is associated with a feature vector having an unknown reward-generating function. We compared already available algorithms like LinUCB, Neural UCB, Neural LinUCB, and Neural Linear. Implemented Neural LinUCB algorithm on the real-world dataset, which has finite action, and modified the algorithm for infinite action cases to optimize an unknown polynomial function. Our algorithm does not use Deep Neural Network to optimize the unknown reward function, as in the case of Neural LinUCB; instead, it relies on the reinforcement learning algorithm only. The objective of algorithm is to learn the maxima of the reward generating function which is a polynomial but degree of polynomial is unknown to the agent. The proposed algorithm is able to learn reward generating function in most of the cases. 2023-06-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/21303 Title: Ensemble-based deep learning architecture for medical image segmentation Authors: Reddy, Malapati Akhil Kumar Abstract: In recent years, skin cancer has overtaken all other types of cancer in frequency and is steadily growing. Skin cells will be harmed by severe and continuous sun burn. Sunburn is a risk factor for skin cell deterioration, yet there as seen in Figure 1.1, there is a decades-long lag time between sunburn and the development of skin lesions. Biomarkers are molecules that, when found or assessed, reveal details about a disease that goes beyond the usual clinical characteristics of a clinician.In the United States, two out of every ten persons suffer from skin diseases, the majority of which are connected to skin cancer (US). The two kinds of skin cancer are melanoma skin cancer (MSC) and non-melanoma skin cancer (NMSC). The four types of malignancies that make up NMSC are depicted in Figure 1.1. Examples of malignancies that affect the skin include Cutaneous Adnexal Carcinomas (CAC), Merkel Cell Carcinoma (MCC), Cutaneous Squamous Cell Carcinoma (CSCC), and Basal Cell Carcinoma (BCC). MSC is the most common malignancy after NMSC, causing 99% of all skin cancer fatalities but only constituting 1% of all skin ma lignancies. In the United States, more than 3.5 million NMSC cases are treated annually. Radiologists who use computer-aided diagnosis in clinical diagnosis can benefit from continuing research on automatic picture segmentation using medical imaging modalities. Its major objective is to replace medical image processing, which is impossible to improve, and to replace it with the infrastructure required for effective clinical diagnosis workflow. Medical image segmentation uses 2D or 3D medical pictures to manually, partially, or fully extract the area of interest (object). It is essential to analyse and evaluate the data set using computers due to its wide range of characteristics and size. It assists in defining the area of interest and provides anatomical details for clinical diagnosis and the benefit of radiologists. 2023-06-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/21302 Title: Breast Cancer Infrared Image Classification Authors: Varshney, Ashish Abstract: Breast cancer is the most typical malignancy among women. Different Researchers have devised procedures all around the world since early diagnosis leads to a better prognosis. Several studies have shown that infrared imaging is an effective test for breast cancer tool. This research provides a method for assessing infrared thermal of the breast using several procedures followed to identify patients' images as healthy or unhealthy due to disease like cancer maligancy. Many approaches, such as Support Vector Machines, rely on handmade features and classical classifiers. Breast cancer diagnosis utilising deep neural networks with attention models and transfer learning will be more accurate than using neural networks alone. Our research intends to assess how well deep learning models that have already been taught perform at identifying breast cancer. We utilise Prewitt and Roberts edge detectors to generate outputs from the raw thermal breast images. The DenseNet121 model receives the original image and these two edge-maps as input.in 3-channel picture form. It includes image preprocessing, transfer learning and deep attention on pre-trained models and visualises the result using grad Cam. Our proposed work improved the accuracy, precision, specificity, sensitivity and recall. 2023-05-01T00:00:00Z http://localhost:8081/jspui/handle/123456789/21301 Title: PORTFOLIO FORMATION BASED ON TECHNICAL INDICATORS FOR INTRADAY TRADING USING DEEP LEARNING TECHNIQUES Authors: Prasanth, Basavala Bhanu Abstract: Portfolio management plays a crucial role in optimizing investment returns while managing risks for both individuals and organizations. With advancements in technology, intelligent models can assist investors and analysts in mitigating investment risks. This study proposes an integrated approach that combines Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) networks with the Black-Litterman model for intraday trading. The study utilizes daily stock data from the Nifty 50 Index spanning from January 2010 to January 2022. The adopted methodology incorporates input features in the form of technical indicators and lagged observations, Principal Component Analysis (PCA) is employed for dimension reduction. The first stage involves predicting the close price of stocks using various deep learning models and ARIMA. Comparisons reveal that the hybrid GRU and LSTM model consistently outperforms other benchmark models across multiple regression metrics for most stocks. The top performing stocks are then selected for further analysis. In the second stage, the Black-Litterman model is utilized to determine the proportional distribution of capital for each individual stock that has been selected, incorporating the predicted results from the GRU_LSTM models as investor views. The performance of the proposed approach is compared against the Modified mean variance model, LSTM model, and equally weighted model. Overall, the hybrid GRU and LSTM model with the Black-Litterman model exhibits superior performance in terms of Sharpe ratio, Maximum Drawdown, Sortino ratio, and Calmar ratio. This research demonstrates the potential of integrating deep learning techniques with statistical portfolio optimization methods for enhanced portfolio management outcomes. 2023-06-01T00:00:00Z