http://repository.iiitd.edu.in/xmlui/handle/123456789/1722 Year-2024 2026-09-20T13:32:52Z http://repository.iiitd.edu.in/xmlui/handle/123456789/2171 Multi-modal sensor fusion for geometric and semantic scene understanding with graphs Rao, Navvrat; Anand, Saket (Advisor) This thesis project, titled Multi-Modal Sensor Fusion for Geometric and Semantic Scene Understanding with Graphs, aims to highlight the constraints inherent in sensors such as LiDAR and cameras and highlight current tools and algorithms for clear interpretation of the environment to occupancy grid. We propose a solution using late fusion for these sen sors, then apply the Kalman filter to the fused data to estimate the detected obstacle’s state. The information derived from this process will then be integrated into the occupancy grid for enhanced accuracy and navigational utility. 2024-04-26T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/2170 Multimodal stream processing for accident detection Singh, Vishal; Shah, Rajiv Ratn (Advisor) This report presents a novel architecture for accident anticipation and localization in autonomous driving, integrating monocular depth-enhanced 3D modeling with large language models (LLMs). The system leverages advanced feature extraction techniques, such as MobileNetv2 and Cascade R-CNN, combined with attention mechanisms to process real-time dashcam video inputs. A dual vision attention mechanism enhances feature representation, while a dynamic object attention mechanism prioritizes high-risk objects. The architecture incorporates a GRU-based accident anticipation module and an attention-driven accident localization module to predict accident probabilities and identify hazardous objects. Finally, the model generates real-time verbal alerts using LLMs, offering contextually relevant warnings to passengers, enhancing safety and aware ness. This hybrid approach aims to improve both accident prediction accuracy and anticipation time, making it suitable for deployment in real-world autonomous driving systems. The model’s effectiveness is evaluated using metrics like average precision (AP) and mean Time to Accident (mTTA). 2024-11-26T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/2169 Development of large language models and tools to study patents and develop insights Vyas, Madhav; Grover, Anuj (Advisor) The development and evolution of Static Random Access Memory (SRAM) has been a cornerstone in the advancement of semiconductor technology, largely due to its ability to deliver high-speed performance while consuming minimal power. SRAM is widely utilized in applications requiring fast and efficient memory access, such as microprocessors, cache memory, and embedded systems. As the demandfor morepowerful, energy-efficient, and scalable memory solutions grows, understanding the underlying innovations and trends within SRAM technology has become increasingly vital. This project presents a novel approach by combining the power of Large Language Models (LLMs) with a chatbot designed specifically for the analysis of SRAM-related patents. The mainobjective of this project is to harness the capabilities of LLMs to create a tool that can provide comprehensive, context-sensitive, and insightful information from SRAM patents. By systematically classifying and analyzing a wide array of patent data, this system is able to cover key aspects of SRAM technology, such as technical specifications, design innovations, architectural variations, performance optimization techniques, manufacturing advancements, and emerging trends in the field. The rich variety of data available in patent documents presents a unique opportunity to identify cutting-edge developments, breakthroughs, and potential challenges facing the SRAMindustry. 2024-11-30T00:00:00Z http://repository.iiitd.edu.in/xmlui/handle/123456789/2165 Comprehensive multi-omics integration for cardiovascular metabolism prediction using machine learning Dhiman, Yash; Ambaprasad, Akshay; Patel, Yaksh; Ray, Arjun (Advisor) CVDs cause more deaths than any other single disease worldwide, with an immense requirement for developing novel strategies in early risk stratification and personalized interventions. Pro viding real-time insights into small-molecule metabolites, metabolomics offers great potential perspectives on cardiovascular metabolism. Integrated in a multi-omics framework along with genomics and proteomics, metabolomics helps identify novel biomarkers and boost the predictive power. Actually, it is through the use of AI, particularly ML and DL, that enables to transform the prediction and diagnosis of CVD risk based on the efficient analysis of layers in complex data. This scoping review clusters literature into ML, DL, and statistical approaches but focuses on hybrid models integrating all these methodologies. Hybrid models that used RFE, CNNs, and statistical validation achieved the highest predictive accuracy, over 93 percent with high sen sitivity and specificity. DL models showed good performance with a mean accuracy of 92.5 percent, performing very well with high-dimensional data and predicting uniformly across dif ferent datasets. ML models enabled the interpretation of features by feature importance analysis that went further in the discovery of biomarkers and clinical utility. This validated the biological relevance of findings obtained, thus providing strong and dependable insights. The risk stratification of CVD and precision medicine are highly advanced by integrating AI driven metabolomics into multi-omics frameworks. These hybrid approaches reach an unprece dented degree of accuracy, interpretability, and clinical relevance with ML, DL, and statistical methods and hold the transformative power for early detection and personalized treatment of cardiovascular care. 2024-01-01T00:00:00Z