Anna Dai

Anna Dai

Researcher in ML/AI

EPFL NLP Lab

Biography

Anna Dai is currently a research assistant at the EPFL NLP Lab, under Professor Antoine Bosselut and Postdoctoral Researcher Syrielle Montariol’s guidance. She has a unique interdisciplinary background in business economics and data science and across industry and academia. She is passionate about responisibly building ML/AI systems for make a positive social impact. Some of her research interests include natural language processing, model security and federated learning.

Interests
  • Ethical ML/AI
  • Natural Language Processing
  • Biomedical Applications
  • Interdisciplinary Research
Education

  • MSc in Interdisciplinary Data Science, 2023

    Duke University

  • BA in Business Economics, 2016

    University of California, Los Angeles (UCLA)

Experience

 
 
 
 
 
EPFL - NLP Lab
Master’s Valorisation Intern (Research Assistant)
EPFL - NLP Lab
September 2023 – Present Lausanne, Switzerland
Developed skill extraction and matching pipeline for job postings, course descriptions, and resumes, leveraging in-context learning through LLMs
 
 
 
 
 
Duke University
Research and Teaching Assistant
Duke University
August 2022 – August 2023 Durham, United States

Research at CEI Lab: Reviewed literature and implemented baseline models from literature on defenses against adversarial model extraction attacks and label protection in the vertical federated learning setting

Teaching at MIDS and Fuqua: Held weekly office hours and graded assignments for Data Engineering Systems, Data Analysis at Scale in the Cloud, and Fraud Analytics courses

 
 
 
 
 
Stitch Fix
Data Science Intern
Stitch Fix
June 2022 – August 2022 San Francisco, United States
Explored unknown problem space of incorporating external fashion trends into our historical-data-based forecasting models
 
 
 
 
 
EY
Senior Tax Consultant, Tax Staff
EY
January 2017 – April 2021 San Francisco, United States
Managed teams of 5-10 members and juggled multiple projects simultaneously to conduct research tax credit studies for clients in the financial and technical sectors
 
 
 
 
 
UCLA
Teaching Assistant
UCLA
September 2014 – June 2016 Los Angeles, United States
Held weekly office hours and proctored/graded exams for three Taxation and Business Law classes during the school years

Publications

ModelGuard: Information-Theoretic Defense Against Model Extraction Attacks
Proposed novel defense against adaptive model extraction attacks through prediction perturbation by leveraging information theory.
Minxue Tang, Anna Dai, Louis DiValentin, Aolin Ding, Amin Hass, Neil Zhenqiang Gong, Yiran Chen, Hai "Helen" Li
August, 2024 USENIX Security 2024
PDF
JobSkape: A Framework for Generating Synthetic Job Postings to Enhance Skill Matching
Developed a framework to generate synthetic job postings using LLMs to address the lack of annotated job posting data that support skill extraction and matching tasks.
Anna Dai, Antoine Magron, Mike Zhang, Syrielle Montariol, Antoine Bosselut
March, 2024 NLP4HR (EACL 2024)
PDF Code Slides
Course Recommender Systems Need to Consider the Job Market
Leveraged LLMs in NLP pipeline to extract skill and match them to a known taxonomy from job postings, course descriptions, and resumes and compared greedy heuristics-based and reinforcement learning-based recommendation systems to recommend courses to job seekers based on their resume and job postings.
Jibril Frej, Anna Dai, Syrielle Montariol, Antoine Bosselut, Tanja Käser
January, 2024 SIGIR 2024 Perspectives
Preprint
PlugVFL: Robust and IP-Protecting Vertical Federated Learning against Unexpected Quitting of Parties
Research on robustness of vertical federated learning convolutional models against performance and IP leakage risks when nactive parties unexpectedly quit during deployment
Jingwei Sun, Zhixu Du, Anna Dai, Saleh Baghersalimi, Alireza Amirshahi, David Atienza, Yiran Chen
March, 2023 preprint
Preprint

Projects

Drug Diversion Detection (Capstone Keynote)
As my master’s capstone project, I worked with my team of four to develop a machine learning model to detect drug diversion by anesthesiologists from the Duke University Hospitals.
Drug Diversion Detection (Capstone Keynote)
Predict the Look @ SFIX
As a member of the Algorithms team at Stitch Fix, I worked on a project to predict fashion trends based on external data sources to better inform our buyers and designers.
Predict the Look @ SFIX

Contact

Please reach out to me if you have any questions or would like to collaborate!