Melanie Subbiah

I'm Melanie Subbiah.

I am a 3rd year Computer Science PhD student, advised in Natural Language Processing by Kathy McKeown at Columbia University. I work on narrative summarization and various aspects of text safety online. Alongside my NLP research, I am interested in AI for climate work and opportunities to speak and write about NLP.

podcast thumbnail

I was interviewed on an episode of the SuperDataScience podcast by Jon Krohn, which has almost 40,000 listens.

> listen

Wired

I spoke about machine learning for a Wired 5 Levels of Difficulty YouTube video, which has over 1.5 million views.

> watch

GPT-3

The GPT-3 paper, which I am a co-lead author on, won a Best Paper Award at NeurIPS 2020. This paper has been cited over 10,000 times.

> read

Education

Columbia

Columbia University

Ph.D. Candidate

Advisor: Kathleen McKeown
Awards:

  • Amazon CAIT PhD Fellowship 2023
  • NSF GRFP Honorable Mention 2021 (Comp/IS/Eng - Natural Language Processing)
  • Best Paper Award at NeurIPS 2020 (Language Models are Few-Shot Learners)
  • Presidential and SEAS Fellowship 2020
Columbia

Columbia University

M.S. in Computer Science

Completed as part of MS/PhD

Williams

Williams College

B.A. in Computer Science, with highest honors

Graduated magna cum laude
Thesis: Using Text Abstraction and LSTM Language Models for Domain-Independent Narrative Generation
Advisor: Andrea Danyluk
Awards:

  • Phi Beta Kappa Commencement Student Speaker
  • Best Thesis Presentation award in Computer Science
  • Magna Cum Laude, Phi Beta Kappa, Sigma Xi
  • Honorable Mention award in short story writing

Work

OpenAI

Meta

Research Intern

Neural interfaces research - deep learning with EMG data.

OpenAI

OpenAI

Member of Technical Staff

Generative language models research - GPT-3 and the OpenAI API.

Apple

Apple

Machine Learning Research Engineer

Autonomous systems research - robotics, RL, computer vision, and NLP.

Research

M Subbiah, A Bhattacharjee, Y Hua, TS Kumarage, H Liu, K McKeown. Detecting Harmful Agendas in News Articles. ArXiv; Jan 2023; Online.  > PDF

S Levy, E Allaway, M Subbiah, L Chilton, D Patton, K McKeown, WY Wang. SafeText: A Benchmark for Exploring Physical Safety in Language Models. Proceedings of EMNLP; Dec 2022; Online.  > PDF

A Mei, A Kabir, S Levy, M Subbiah, E Allaway, J Judge, D Patton, B Bimber, K McKeown, WY Wang. Mitigating Covertly Unsafe Text within Natural Language Systems. Findings of EMNLP; Dec 2022; Online.  > PDF

M Subbiah, K McKeown. Understanding Identity Signalling in Persuasive Online Text. Presented at ICWSM, International Workshop on Social Sensing (talk and vision abstract); Jun 2021; Online.  > PDF

Co-organizer of ICLR Workshop on Enormous Language Models. 2021.  > workshop website

Best Paper Award
TB Brown*, B Mann*, N Ryder*, M Subbiah*, et al. Language Models are Few-Shot Learners. Neural Information Processing Systems; Dec 2020; Montreal, CA.  > PDF

M Subbiah, et al. Augmenting Training Data with Simulated Images. Presented at Neural Information Processing Systems, Women in Machine Learning Workshop (poster); Dec 2018; Montreal, CA.

M Maher, M Subbiah, N Apostoloff. Cascaded Dataset QA. Presented at Neural Information Processing Systems, Women in Machine Learning Workshop (talk and poster); Dec 2018; Montreal, CA.

UC Nygaard, Z Li, T Palys, B Jackson, M Subbiah, M Malipatlolla, V Sampath, H Maecker, MR Karagas, KC Nadeau. Cord blood T cell subpopulations and associations with maternal cadmium and arsenic exposures. PLoS One. 2017 Jun 29;12(6):e0179606. doi: 10.1371/journal.pone.0179606. PMID: 28662050; PMCID: PMC5491028.  > PDF

Speaking

PhD Candidacy Exam: Narrative Summarization. 2023.  > Slides

Pearson Publishing's AI Catalyst Conference: NLP with ChatGPT. 2023.  > Event

SuperDataScience podcast: GPT-3 for Natural Language Processing. 2022.  > Podcast

Wired 5 Levels of Difficulty video: Machine Learning. YouTube. 2021.  > Video

Invited seminar talks on GPT-3 at Stanford University and New York University. 2020.  > Slides

Invited Computer Science colloquium talk at Williams College. 2020.

Williams College Commencement Phi Beta Kappa speaker. 2017.  > Video

Teaching

  • TA (and lecture) for Natural Language Generation and Summarization at Columbia
  • CS Tutor for Columbia Athletics
  • TA for Intro CS and Data Structures at Williams

Mentorship

  • Williams College CS undergrad mentor
  • Lumiere Education research mentor
  • OpenAI Scholars program development
  • OpenAI Scholars mentor
  • Institute of International Education TechWomen mentor
  • Williams College Underrepresented Identities in CS leader

Interests

  • Natural Language Processing
  • Language Models
  • Reinforcement Learning
  • Deep Learning
  • Computational Creativity
  • AI for Climate
  • AI for Energy Efficiency

Contact

melanie.s.subbiah@gmail.com

I enjoy creating AI systems that are capable of creative behavior and problem solving. I value ethical and sustainable approaches to AI research.