AI Research Scientist, Multimodal Generation

Meta•Menlo Park, CA

About The Position

Meta is seeking an AI Research Scientist to join our Multimodal Generation Research team. We are looking for recognized experts in media (image or video or audio) generation models to work in areas like vision encoders, data filtering/curation for pre and post-training, RL. Seeking a candidate who will have an interest in producing and applying new science/systems/technologies to help us develop media generation models and bringing the latest research to Meta products for connecting billions of users. They will work with an interdisciplinary team of scientists, engineers, and cross-functional partners, and will have access to cutting edge technology, resources, and research facilities.

Requirements

Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience
PhD in Computer Science, Machine Learning, or a relevant technical field
Practical experience with pre-training, mid-training or SFT data curation for large foundational models and experience working with organic, synthetic, agentic, or reasoning data for Multimodal LLMs
Direct experience in Generative AI and LLM research
Programming experience in Python and hands-on experience with frameworks such as PyTorch

Nice To Haves

First-authored publications at peer-reviewed conferences (e.g. CVPR, NeurIPS, ICCV, ECCV, ACL)
Experience collaborating in cross-functional teams, including product, engineering, and research

Responsibilities

Develop algorithms based on state-of-the-art machine learning and neural network methodologies
Advance our understanding of data research, such as how to overcome data walls and how best to create synthetic data
Post-train foundation models using techniques such as Supervised Fine-Tuning (SFT), Reinforcement Learning from Human Feedback (RLHF), Direct Preference Optimization (DPO), and Low-Rank Adaptation (LoRA)
Work towards long-term research/development goals, while identifying intermediate milestones
Conduct research that enables learning the semantics of data across multiple modalities (audio, images, video, text, and other modalities)
Prioritize research that can be applied to Meta's product development