Senior AI Research Scientist/Specialist, Foundation Models

Merck KGaA Darmstadt Germany - Cambridge, MA

Full-time - Mid Level

Onsite - Cambridge, MA

Chemical Manufacturing

About the position

The AI Research Scientist, Foundation Models role focuses on advancing the understanding of complex diseases through the development and application of biological foundation models. This position involves pre-training and fine-tuning models, analyzing their performance, and collaborating with a cross-functional team to support innovative therapeutic strategies.

Responsibilities

• Collaborate with cross-functional teams to identify research questions and data requirements and develop appropriate solutions.
• Develop and train transformer-based (and related state-space models) foundation models for -omics data.
• Interpret and post-hoc analyze pre-trained models.
• Rigorously benchmark and evaluate the performance of both in-house and publicly available models.
• Host and serve in-house models and make them accessible to scientists across the company.
• Stay up to date with the latest advancements in machine learning and statistics and apply relevant advancements to improve existing methodologies and models.
• Publish research findings in relevant conferences and journals and actively contribute to the scientific community through knowledge sharing and collaborations.

Requirements

• PhD, MS, or BS in Computer Science, Statistics, Physics, or a related field with varying years of experience based on degree.
• Expertise in machine learning and in training, evaluating, and debugging models and data at scale.
• Excellent software design and development skills and strong proficiency in Python.
• Experience with standard deep learning frameworks like PyTorch and the Huggingface ecosystem for working with transformer-based foundation models.
• Excellent communication skills and ability to work collaboratively in a multi-disciplinary team.
• Interest in life sciences problems and disease biology, and willingness to learn from and teach others.

Nice-to-haves

• Demonstrated experience working with models that require multiple GPUs for training and inference.
• Relevant publications in scientific journals and experience contributing to research communities, including NeurIPS, ICML, ICLR, etc.
• Experience with pre-training and multi-modal training of (biological or otherwise) foundation models is a strong plus.
• Familiarity with biological data and previous experience with protein language models and foundation models for omics is a strong plus.

Benefits

• Hybrid work model with three days on-site per week and one remote working day.
• VISA sponsorship available.
• Relocation assistance for domestic/international moves.