Megan Wei
I’m a CS PhD candidate at Brown advised by Professor Ellie Pavlick. I am interested in building more interpretable and controllable multimodal foundation models, providing intuitive methods to empower human creative expression.
I completed my BS and MEng in CS at MIT, where I worked with Professor Josh Tenenbaum on reasoning, creativity, and planning in LLMs and Professor Antonio Torralba on building multimodal benchmarks for embodied agents.
Recently, I’ve worked at Adobe Research on creating models that better understand paralinguistics (prosody, emotion) in speech. Previously, I held software, product, and research roles at Microsoft, IBM, and several startups.
Recent News
- Organized the WiML Social @ COLM 2026 in SF!
- Presented my paper Do Speech Emphasis Models Generalize across Languages and Emotions? at Interspeech 2026.
- Interviewed on: How does AI impact our cognitive abilities?
- Working at Adobe Research Speech AI Team in SF!
- Gave an invited talk at UCSD MUSAIC Group on Do Music Generation Models Encode Music Theory?
- Gave a talk on my paper Do Music Generation Models Encode Music Theory? at ISMIR 2024
- Gave an invited talk at Boston AI Music Meetup on Do Music Generation Models Encode Music Theory?
- Gave an oral presentation (spotlight) at BayLearn 2024 on Do Music Generation Models Encode Music Theory?
- Working at Adobe Research Speech AI Team in Seattle!
- Presented my poster on Do Music Generation Models Encode Music Theory? at New England NLP 2024
- Presented our group poster at NYC Computer Vision Day 2024
Publications
-
Activation Steering the Nan-Shan in Typhoon
-
Ownership in AI-Assisted Everyday Tasks
-
Harmonic Representations in Symbolic Music Transformers are Localized and Controllable
-
Do Speech Emphasis Models Generalize across Languages and Emotions?
-
You May Speak Freely: Improving the Fine-Grained Visual Recognition Capabilities of Multimodal Large Language Models with Answer Extraction
-
Language-Informed Synthesis of Rational Agent Models for Grounded Theory-of-Mind Reasoning On-The-Fly
-
Do Music Generation Models Encode Music Theory?
-
The Neuro-Symbolic Inverse Planning Engine (NIPE): Modeling Probabilistic Social Inferences from Linguistic Inputs
-
Structured, flexible, and robust: benchmarking and improving large language models towards more human-like behavior in out-of-distribution reasoning tasks