Position SummaryAs a Research Scientist in the Vision Language Model (VLM) team, your role will be central to advancing state-of-the-art multimodal foundation models that integrate visual understanding, reasoning, and agentic capabilities. You will work on the research and development of large-scale VLM systems, spanning model architectures, data recipes for pre-training and post-training, and evaluation benchmarks. The role combines cutting-edge research with practical engineering, emphasizing large-scale data processing, filtering, and weighting pipelines, distributed training systems, and reinforcement learning algorithms and systems for multimodal reasoning and agent development.
Key Responsibilities- Research and development of next-generation Vision Language Models across pre-training, instruction tuning, reasoning, and agents.
>- Develop novel architectures and training methodologies for integrating visual understanding, language reasoning, and tool-use capabilities.
>- Research efficient multimodal learning techniques, including data-efficient training, long-context modeling, model modularity, and inference optimization.
>- Build and improve large-scale multimodal datasets, synthetic data generation pipelines, and evaluation benchmarks for VLM capabilities.
>- Investigate multimodal reasoning, agentic behavior, OCR, grounding, document understanding, chart understanding, and visual question answering capabilities.
>- Contribute to technical reports, research publications, and open-source software.
>- Represent MBZUAI at research conferences and industry events, showcasing advancements in multimodal foundation models and large-scale AI systems.
>- Mentor junior researchers and collaborate across teams to drive impactful research initiatives.
>
Academic QualificationsPhD or equivalent research experience in Machine Learning, Computer Vision, Natural Language Processing, or Multimodal AI.
$150,000 - $450,000 a year
Salary Range
The posted salary range represents the company's good faith estimate of the compensation for this position upon hire. The actual compensation offered may vary within this range depending on individual qualifications, including but not limited to relevant skills, experience, education, certifications, geographic location, and specific business needs.
Professional ExperienceMinimum - Experience working with large language models and/or vision-language models, including pre-training, fine-tuning, evaluation, or inference.
>- Strong Python and PyTorch development skills for large-scale machine learning research.
>- Experience with distributed training systems and large-scale model optimization.
>- Familiarity with multimodal datasets and data processing pipelines involving images, text, and video.
>- Understanding of modern deep learning architectures, including Transformers, attention mechanisms, and multimodal fusion techniques.
>- Experience with ML infrastructure, including model evaluation, debugging, optimization, and large-scale experimentation.
>- Problem-solving and research skills with the ability to independently drive research/engineering projects.
>- Effective communication and collaboration skills for working across research and engineering teams.
>
Preferred Skills- Hands-on experience training or fine-tuning large Vision Language Models or multimodal foundation models at scale.
>- Experience with distributed learning frameworks and infrastructure such as PyTorch Distributed, Megatron, Triton, or CUDA.
>- Research experience in multimodal reasoning, agentic systems, tool use, OCR, grounding, document understanding, or multimodal retrieval.
>- Experience with synthetic data generation, multimodal data curation, or automated evaluation frameworks for VLMs.
>- Familiarity with efficient training and inference techniques such as FlashAttention, quantization, tensor parallelism, pipeline parallelism, or memory optimization.
>- Experience contributing to open-source ML software and large-scale research codebases.
>- Strong publication record in leading AI conferences such as NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, ACL, EMNLP, or related venues.
>- Experience collaborating across research, infrastructure, and product-oriented teams to deliver state-of-the-art multimodal systems.
>
