Furong(Flora) Jia

Publication

• I2I: Initializing Adapters with Improvised Knowledge Tejas Srinivasan, Furong Jia, Mohammad Rostami, Jesse Thomason.
  Conference on Lifelong Learning Agents (CoLLAs) 2023
• GPT4MTS: Prompt-based Large Language Model for Multimodal Time-series Forecasting. Furong Jia, Kevin Zheng, Yixiang Zheng, Defu Cao, Yan Liu.
  The 14th Symposium on Educational Advances in Artificial Intelligence (EAAI-24)
• TEMPO: Prompt-based Generative Pre-trained Transformer for Time Series Forecasting Defu Cao, Furong Jia, Sercan O Arik, Tomas Pfister, Yixiang Zheng, Wen Ye, Yan Liu.
  The Twelfth International Conference on Learning Representations (ICLR 2024)

Education

Research Experience

Projects & Contributions:

1. Prompt-based Large Language Model for Multimodal Time-series Forecasting
   • Established a pipeline for proposing multimodal time series dataset to improve time series forecasting lending the ability of Large Language Models (LLM), leading to a first-authored paper accpeted to EAAI-24.
   • Pre-processed, analyzed, and established a new multimodal time series dataset on the news using GDELT database and incorporated textual information from related news articles through LLM models.
   • Compared state-of-the-art (SOTA) time series forecasting models, and proposed our model (GPT4MTS) by treating textual information as prompt to improve performance.
2. Large Language Models (LLMs) for Time Series
   • Collaborating with Ph.D. students to assess the potential of utilizing Large Language Models (LLMs) for time series tasks through prompting and decomposition, aiming for building foundation model lending LLM.
   • Introduced and employed selection-based prompts to enhance distribution adaption in non-stationary time series.
   • Actively engaged in the complete research cycle: from initial architecture design and implementation to experimentation, visualization, in-depth analysis, and paper drafting.
3. Causal Benchmark in Treatment Effect Estimation
   • Collaborating with Ph.D. students to examine the influence of causal components on current models for treatment effect estimation.
   • Researched and implemented dataset under static setting, experimented and analyzed the effect of the causal components of various models, drafted and compiled findings aiming for a conference submission.

Projects & Contributions:

1. Initializing Adapters with Improvised Knowledge for Multimodal Continual Learning
   • Collaborated with Ph.D. students on a multi-modal continual learning project to enable parameter- efficient knowledge transfer through adapters on question answering tasks.
   • Implemented multiple vision-language tasks, including GQA, COCO-QA, AQUA, trained and imple- mented model architectures to test various settings to enhance forward transfer and reduce knowledge forgetting, resulting in a co-author publication presented at CoLLAs 2023.
2. Prompting Methods for Multimodal Continual Learning
   • Initiated and currently leading a project exploring the use of prompting methods for visual question answering tasks in multi-modal continual learning.
   • Proposed, executed, and analyzed various methods, actively working towards a conference submission to share the findings.

Projects & Contributions:

• ASD Behavior Recognition Project
  • Collaborated with Ph.D. students to develop a large-scale, real-scenario dataset for recognizing children’s behaviors, receiving Spring 2022 Provost’s Research Fellowship.
  • Preprocessed, annotated, and detailed children-robot interactions within the dataset, trained and applied state-of-the-art (SOTA) action recognition models to the dataset, with a focus on refining and comparing various sampling methods.

• Developed, updated and migrated Interaction Lab Website, and professor’s personal website by using Vue.js for front end and Firebase for back end.

Work Experience

• Independently designed, developed, and tested micro-services in Flask and Python to support authentication and retrieve account information from database, and generated scripts for Continuous Integration through Gitlab
• Independently designed, developed, and tested account-information-component that create web components to display account information card using lit-element framework in HTML/CSS and Typescript

• Developed homepages, login, and signup pages independently through HTML/CSS, Angular.js, and Ionic framework to assist the front end development of an internally used mobile application
• Utilized Gitee to contribute to the development and modification of the “Shanghai Hezhang” project using HTML/CSS, Angular.js, and Node.js: setting up service, modifying layout, and adding new functions such as loading animation and notification for system update

Projects

Python, Natural Language Processing, Hugging Face, Large Language Models

• Developed an AI Chatbot for ANRG labs aimed at answering questions regarding lab history, members, research papers, and projects, providing a quick and efficient way for students and recruiters to learn about lab.
• Implemented a Retrieval-Augmented Generation framework to efficiently process lab-specific information through utilizing OpenAI API for query handling and related information retrieval.

Demo Page

Python, Natural Language Processing, Hugging Face, tkinter

• Researched on the current gap of text meme prediction, selectively implemented and trained SOTA model (RoBERTa) on sentiment analysis over GoEmotion dataset, collaboratively developed pipeline for suggesting top-K memes from user’s meme library based on current text, and implemented GUI using tkinter package under Python

Github Repo

HTML/CSS, Javascript, Vue.js, Axios, Webpack, Firebase, Java Spring boot, README

• Cofounder of Ctrl-F, a web platform that bridges students and faculties over research and internship opportunities to cover the blank of similar products in the current market
• Developed all Front-end pages through HTML, CSS, Javascript, Vue.js, enabled user authentication using Firebase, connected front-end with back-end through Axios and Webpack, and deployed through AWS
• Collected professors’ and students’ feedback over the platform to design and develop new functionalities: adding google sign in/up options, allowing users editing own profile and adding affiliated organizations , enabling fast and global searching with multi-layer filters by implementing Elasticsearch

HTML/CSS, Javascript, Vue.js, Firebase, Axios, Webpack, Java Spring boot, AWS

• Cooperatively developed 2BotOrNot2Bot, a web platform to test chat bots for companies by pairing users with chat bots for Turing test
• Developed all Front-end pages through HTML, CSS, Javascript, Vue.js, enabled user authentication using Firebase, and connected front-end with back-end through Axois and Webpack
• Applied chat bots API to 2BotOrNot2Bot using Java Spring Boot code and deployed the web platform through AWS

HTML/CSS, Javascript, Python, Django, Web scraping, Heroku

• Developed and Deployed SCourse, a web platform implemented through Python with Django framework which generates a full schedule according to students’ input course numbers and their preferences of starting time and rate of professor
• Gathered course information from USC class website and then implemented course arrangement for lectures, discussions, and quizzes to display the most prioritized result

Skills