Profile Picture

Zora (Zhiruo) Wang

็Ž‹่Šท่‹ฅ

PhD Student

Carnegie Mellon University

Language Technologies Institute

About Me

I am currently a 2nd year PhD student at the Language Technologies Institute at Carnegie Mellon University working with Daniel Fried and Graham Neubig. My primary research interest is to use programmatic approaches to solve real-world tasks. Particularly, I am working on:

Memory- and Skill-Adaptive Agents
  1. Digital agents that learns navigation workflows in memory [AWM] and specialized skills as actuators [ASI] (coming soon!); build verifiable and efficient toolboxes for math, data analysis, and visual reasoning tasks [TroVE]
  2. These agents often facilitate human verification [TroVE] and improves task success [CowPilot]
  3. On versatile tasks such as presentation making [AutoPresent], software engineering [ODEX], and popular professions [TAC]
Augmented Language Models
  1. Tools [Survey]: designed by human experts [AutoPresent] or LMs [TroVE]; their behaviors to under-specified queries and execution failures [Fail-TaLMs]
  2. Programs: Challenging benchmarks for open domain [ODEX], multilingual [MCoNaLa], and data science [HiTab] problems; Performant code LMs [StarCoder] [ECCO] with [Code RAG], assistive [APIs], and libraries in [DocPrompting]
  3. Knowledge: retrieve texts [FilCo] [ReAtt] [RAGGED] and structured data [WikiTable] [TUTA]

News

๐Ÿพ Older News
  • May 2024: Organized our CMU Agent Workshop ๐Ÿค– with plenty of events -- insightful tutorials, talks, and posters! I also gave two (short) tutorials about tool-augmented LMs and codegen testbeds.
  • Mar 2024: Gave a guest lecture about "Language Agents and Tool Use" at the Advanced NLP course (11-711) course, check out the recordings!
  • Mar 2024: Gave a talk at the [FLAME_ (Foundation and LAnguage Model) ๐Ÿ”ฅ] seminar, about [our recent survey] and [TroVE]
  • Feb 2024: Gave a talk about Language Models with Tools at the LLM as Agent Seminar, about [TroVE] and works in progress ๐Ÿคซ
  • Feb 2024: Gave a lecture about Evaluation (metrics and benchmarks) for the Neural Code Generation (11-891) course ๐Ÿ’ป
  • Jan 2024: TAing for the new course [11-891 Neural Code Generation], reach out if you want to discuss more project ideas ๐Ÿช„
  • Nov 2023: Gave a guest lecture about Evaluation and Benchmarks for Code Generation for the Advanced NLP course (11-711) ๐Ÿ‘ฉโ€๐Ÿซ more details [here]
  • Aug 2023: A talk about ๐Ÿ› ๏ธ Tool using, learning, and making with LLMs at Code Generation Reading Group, check out the [video]
  • Apr 2023: Gave a talk about [ODEX] at the Machine Learning Methods in Software Engineering (video) hosted by JetBrains Research Team ๐Ÿ‘ฉโ€๐Ÿ’ป

    • Upcoming Events

    My Recent Favorite Publications

    Agent Workflow Memory

    Agent Workflow Memory

    Survey on Language Models

    What Are Tools Anyway? A Survey from the Language Model Perspective

    TroVE Took Making

    TroVE: Inducing Verifiable and Efficient Toolboxes for Solving Programmatic Tasks

    Get Connected

    If you want to get connected, discuss potential project ideas, ask about CMU application, or chat about any other relevant topics, my office hours are available every Friday 4-5 pm EST. Email me to secure a time to chat!

    I often mentor a small number of students every semester, please fill out this application form if you're interested in working with me! However, as I got plenty of emails, I might not be able to reply to all of them.

    If you are from underrepresented groups, or do not have much research experience, you are encouraged to reach out!

    If You're Interested in My Name

    My name in Chinese is ็Ž‹่Šท่‹ฅ, which reads as Wang, Zhiruo. It is usually hard for non-native speakers to pronounce, so you can also call me Zora (as ZR is similar to Zhi Ruo).

    I love my name, especially in Chinese characters, since it has a more beautiful meaning than in English alphabets. ่Šท stands for ็™ฝ่Šท (Angelica dahurica) and ่‹ฅ stands for ๆœ่‹ฅ (Pollia japonica), which are two kinds of Chinese herbal medicine. Also, ่Šท่‹ฅ is a beautiful vanilla ๐ŸŒฟ