Research Scientist Intern, Foundation Model Optimization (PhD)

Meta's mission is to give people the power to build community and bring the world closer together, and the Meta Reality Labs organization focuses on delivering this vision through AR/VR/SG devices. Meta's Efficiency and On-Device AI team within the Reality Labs organization is seeking a Research Scientist Intern with a focus on language/vision model optimization. As a Research Scientist Intern, your primary objective is to develop cutting-edge efficient LLMs and vision foundation models.

Responsibilities

• Perform LLM and vision foundation model research that leads to publications in top-tier CV/NLP/ML conferences.
• Develop model optimization algorithms and infrastructure to enable deployment of models in resource-constrained settings, such as mobile phones and AR/VR/MR/SG devices.
• Research joint optimization of models and the surrounding assistant system, including planning, memory, tool use, and coordination across device and cloud.
• Explore personalization, model compression and efficiency, and novel model architectures including omni-modal capabilities.
• Collaborate with cross-disciplinary research and engineering teams.

Qualifications

• Currently is in the process of obtaining a PhD in the field of Artificial Intelligence or related field
• Must obtain work authorization in the country of employment at the time of hire, and maintain ongoing work authorization during employment
• Hands-on experience in deep learning algorithms and techniques, e.g., transformers
• Experience with deep learning software libraries such as PyTorch/JAX
• Experience with scripting languages such as Python and shell scripts Intent to return to degree-program after the completion of the internship
• Experience with large scale foundation model training and on-device foundation model optimization
• Experience in one or more of: LLM agents and orchestration, personalization, model compression and efficient inference, efficient model architectures, multi-modal modeling, data-centric approaches for model tuning and evaluation
• Proven track record of publications at CV/NLP/ML conferences/journals such as NeurIPS, ICLR, ICML, CVPR, ICCV, ECCV, AAAI, ACL, EMNLP