School of Engineering Department of Computer Science and Engineering 111 Foundation Model-as-a-Service at Edge Supervisor: GUO Song / CSE Student: GUPTA Harsh Vardhan / CPEG Course: UROP 1000, Summer The Large Language and Vision Assistant (LLaVA) is a multi-modal large language model (MLLM) that extends traditional large language models (LLMs) by integrating image processing with textual data to better understand user prompts. In this UROP project, we explored the architecture of LLaVA, independently reproduced its fine-tuning process, and conducted a comparative case study of the reproduced model against GPT-4 and Microsoft Copilot. Through human evaluations we performed, LLaVA demonstrated accurate responses, though they lacked detail. This work aims to provide a deeper insight into the architecture and implementation of the LLaVA model and MLLMs in general. Open Topic in Algorithms and Complexity Supervisor: KAFSHDAR GOHARSHADY Amir / CSE Student: REHMAN Abdul / COMP Course: UROP 1000, Summer Finding polynomial-time solutions for problems is challenging unless it can be shown that = . However, this difficulty applies to general solutions. By limiting the problem definition, such as restricting graphs to be trees or vertices to have a degree at most , we can create simpler cases where polynomialtime algorithms can be applied to solve the problem. This concept of restricting the problem, known as parameterization, has been extensively studied to provide feasible solutions for problems. In this report, we present various parameterization techniques in the context of finding a -sized vertex cover of a graph. We explore methods such as kernelization, bounded search trees, linear programming, crown decomposition, and approximation algorithms, aiming to provide a comprehensive understanding of how to tackle an problems in a more tractable manner.
RkJQdWJsaXNoZXIy NDk5Njg=