Conditional Generative Modelling Internship

About Valence Labs

Valence Labs is an AI research engine within Recursion dedicated to industrializing scientific discovery to radically improve lives. Combining the intellectual freedom of academia with the resources and stability of industry, our focus is the development of highly-autonomous systems that will spearhead a fundamental shift in the way treatments are discovered and developed for complex disease. Our research is driven by optimism, purpose, and a shared vision for a healthier tomorrow. We publish in top journals and conferences, are deeply committed to open-science and open-source, and maintain some of the largest and most active research communities in our industry. Our team is located in London and Montreal, where we share close connections with Mila, the world’s largest deep learning research institute.

About the Role

We’re seeking motivated interns to join our Extrapolation Unit and contribute to groundbreaking research in conditional generative modelling for biological applications. This position focuses on developing and adapting advanced image-editing techniques to simulate biological perturbations, such as the effects of novel drugs or gene knockouts. We are looking for individuals with strong engineering skills, including expertise in designing, implementing, improving, and deploying distributed machine learning systems at scale. In addition, we highly value proficiency with state-of-the-art generative modeling techniques, especially diffusion-based and adversarial approaches, as well as exceptional problem-solving skills.

In this role, you will:

• Investigate and implement generative modeling techniques, including diffusion- and adversarial-based approaches, for conditional image editing and unpaired translation.
• Adapt these techniques to simulate biological perturbations, such as drug applications or gene knockouts.
• Formulate and train large-scale models using both public datasets (e.g., RXRX3) and our larger internal datasets to model biological effects on cell images.
• Support Valence Labs' research agenda towards creating a "causal digital twin" or "virtual cell" to accurately simulate cell perturbations.
• Collaborate with an interdisciplinary team of machine learning experts and biological scientists to refine models and approaches.
• Present and communicate research findings through talks, blog posts, publications, and conferences.

A successful candidate will have most of the following:

• Currently enrolled in a post-doctoral fellowship, PhD, or Master's degree program.
• Strong programming skills and understanding of modern software development practices, especially in Python.
• Proven track record in machine learning, including designing new architectures, hands-on experimentation, analysis, visualization, and model deployment.
• Experience with generative modeling techniques, particularly diffusion-based or adversarial methods for applications in image processing.
• Demonstrated capability to understand and summarize scientific content and implement deep learning models based on descriptions from publications.
• Strong knowledge of linear algebra, calculus, and statistics.
• Passion for applying ML research to real-world problems, especially in the life sciences.

Nice to have:

• Authorship of a publication in peer-reviewed conferences (e.g., NeurIPS, ICML, ICLR, or similar).
• Contribution to high-visibility ML codebases.
• Experience training large-scale models in an industry setting.
• Scientific knowledge of biology, particularly in cellular biology, along with experience working in a scientific research environment across disciplines.

Valence Labs is committed to creating a diverse and inclusive environment, where understanding and accommodating personal needs and preferences is a priority. Join our multidisciplinary team of passionate researchers, eager to push the boundaries of ML research and contribute to industrializing scientific discovery to radically improve lives.