Research Scientist Intern - Post-Training (RLHF)

Jasper is the leading AI marketing platform, enabling the world's most innovative companies to reimagine their end-to-end marketing workflows and drive higher ROI through increased brand consistency, efficiency, and personalization at scale.

Jasper has been recognized as "one of the Top 15 Most Innovative AI Companies of 2024" by Fast Company and is trusted by nearly 20% of the Fortune 500 – including Prudential, Ulta Beauty, and Wayfair. Founded in 2021, Jasper is a remote-first organization with team members across the US, France, and Australia.

Jasper Research is seeking a highly motivated intern to advance the frontiers of open source image-generation and image-editing applications. In this role, you will be instrumental in developing new state-of-the-art open-source text-to-image and image editing models while collaborating closely with our talented team of researchers and engineers. The internship duration is 6 months.

We have historically relied on third-party existing open-source image foundational models for its core applications (replace-background, image uncropping, image relighting …), mastering model fine-tuning methods. One of the team’s current core projects consists now of building our own open-weights foundational text-to-image model from scratch.

This role is open to candidates located in Paris. It will be a hybrid setup, which requires you to come into the office when necessary.

What you will do at Jasper

As a Research Scientist Intern, you will work closely with our research team to design the best-suited RLHF techniques for diffusion models. This role offers a unique opportunity to contribute to the development of a foundational open-source model, addressing challenges in scalability, fidelity, and generalization. You will engage in both theoretical and applied research, collaborating with experts in machine learning, computer vision, and natural language processing. The main goals of this internship are

1. Research & Development: Conduct literature reviews, propose and implement innovative methods to fine-tune text-to-image models with reinforcement learning methods (RLHF, DPO)

2. Model Training & Evaluation: Participate in the fine-tuning stages of the training of large-scale text-to-image models, conduct rigorous ablations, design evaluation metrics, and analyze model performance.

3. Documentation & Communication: Document research findings, prepare technical reports, and participate in the external communication of the results.

4. Open Source & Community: Contribute to an ambitious open-source project, publish research findings, and engage with the broader AI community.

• Currently enrolled in a Ph.D. or M.Sc program in Machine Learning, applied mathematics or computer science (Ph.D, preferred).

• Experience with RLHF techniques with application to either Large Language Models or Diffusion Models.

• A genuine interest in the field and a strong motivation to contribute to open-source initiatives, with a potentially proven track record through personal projects or previous experience in deep learning, especially generative models (e.g., diffusion models, GANs, VAEs, transformers).

• Strong coding abilities in Python and deep learning frameworks (PyTorch, TensorFlow, JAX).

• The ideal candidate will possess a strong critical thinking and problem-solving mindset, coupled with excellent teamwork skills.

• Being available for period of 6 months.

Nice to have

• Strong programming Python skills, including software engineering best practices to produce high-quality code.

• Experience with distributed training and large-scale systems on GPU clusters

• Experience with large-scale data processing

• Contributions to open-source projects.

• Proven track record of achieving significant results as demonstrated by first-authored publications in major conferences and journals such as CVPR, ECCV, ICCV, ICLR, NeurIPS

• Flexible, hybrid work environment. Our office is based at Station F in Paris, the vibrant hub of the French startup ecosystem. Our efficient and lean team at Station F thrives on innovation and collaboration.

• Competitive compensation package