School of Engineering Department of Computer Science and Engineering 115 Interaction Design for Human-AI Collaboration Supervisor: MA Xiaojuan / CSE Student: HE Jinwen / COGBM Course: UROP 1100, Spring This project aims to gain insight into human-computer interaction and improve an agent's capability to learn a new task. The project has an objective but is still in progress at the preparation stage in search of methods that can satisfy the objective. For now, we have identified the approach to use data generated from humans interacting with robotic arms in virtual reality to perform Visuomotor Policy Learning via Action Diffusion (Chi et al., 2023) to generate robot actions that align with human value. This progress report will introduce the progress on methods mentioned above and a literature review on several papers related to human value verification that could be further examined to add to the project. Interaction Design for Human-AI Collaboration Supervisor: MA Xiaojuan / CSE Student: ZHAO Bolin / DSCT Course: UROP 1100, Fall Langchain is a Python library that provides easily applicable interfaces for working with OpenAI's powerful large language models (LLMs). This report covers the intuitions of basic functions of the Langchain library, including Output Parsers, Prompt template, and memory storage, with specific coding examples. Among them, output parsers are used to arrange output information to the user-intended forms; prompt templates are templates for prompts used to provide a structured input format for language models. Additionally, the report covers the execution of the paper implementing Game of 24. The implementation code of this paper involves the usage of BFS algorithms, greedy algorithms, and tree-of-thoughts prompting methods to interact with OpenAI's language models and generate solutions for the game. Compilation, Optimization, and Metaprogramming Supervisor: PARREAUX Lionel / CSE Student: LAU Yan Hei / COMP Course: UROP 1100, Fall UROP 3200, Spring Defunctionalization is a compile-time transformation which takes higher-order function calls and replaces them with a first-order apply function. It is used as a compilation technique in many programming languages. In this project, we implement defunctionalization using the Simple-sub algorithm as the core of the program analysis.
RkJQdWJsaXNoZXIy NDk5Njg=