School of Engineering Department of Computer Science and Engineering 122 Large Language Models as Your Machine Learning Experts Supervisor: DI Shimin / CSE Student: JIA Yusheng / COMP Course: UROP 1100, Fall During this semester, I read several papers on the implementation of AutoML with LLM, which provided me with a preliminary understanding of this field. As a novice in artificial intelligence, I gained insight into several novel concepts through the review and subsequent searches for related literature. While the subject matter was complex, I found the experience intellectually stimulating. In this report, I will mainly discuss my understanding of these fundamental theories, as well as my harvest through this process. Large Language Models as Your Machine Learning Experts Supervisor: DI Shimin / CSE Student: LI Haowei / COMP Course: UROP 2100, Fall This report presents a comprehensive overview of my UROP2100 research experience. The propose of this project is to explore and leverage the power of Large Language Models (LLMs), which aims to integrate with AutoML techniques to perform a more efficient way to automate Machine Learning on Graphs. In this semester I focused on the construction of the general graph neural network (GNN) pipeline, which help automate the GNN model selection process. This report will cover the introduction to the general architectures of the pipeline, while I will also discuss on the part of my work which focus on the improving the features that the pipeline can provide for handling variant receptive fields required in different graphs. Hence we can achieve a more general and efficient GNN model selection process. The first section is a brief introduction to the general graph neural network pipeline architecture, which in our cases we build upon the powerful GraphGym framework. The second section will discuss the additional features that I have added to the pipeline. We will also present the experimental results to evaluate the performance of the proposed features. The last section will be discussion part, where I will discuss the potential future work and the possible improvements that can be made by integrating this powerful pipeline for the future research. Large Language Models as Your Machine Learning Experts Supervisor: DI Shimin / CSE Student: LI Yifan / COMP Course: UROP 1100, Fall This progress report explores integrating Automated Machine Learning with Large Language Models through knowledge-driven methods. LLMs has shown remarkable potential in automating tasks like pipeline design and architecture optimization, which significantly enhance AutoML frameworks efficiency. Our group’s previous research focused towards automating Graph Neural Network(GNN) design using a knowledgebased approach. This methodology demonstrated significant reduction in computational requirements while making the process more straightforward for practitioners. Building upon these findings, we propose further exploration of knowledge-driven techniques. The ultimate goal is to improve the adaptability and scalability of AutoML systems across various domain applications.
RkJQdWJsaXNoZXIy NDk5Njg=