About this role
Summary: Leads offline reinforcement learning research from initial questions through shared findings, building on prior policy work and evaluating learned control policies on internal and physical systems.
About the company
Our client is a technology company.
The role
Raydar is recruiting for this role on behalf of our client. Lead a research agenda in offline reinforcement learning, taking ideas from initial question through to shared findings. The position prizes scientific depth and judgment over engineering breadth, and pairs you with a more established colleague while leaving space to chart your own path.
What you'll do
- Drive offline reinforcement learning research from start to finish, including choosing which questions to pursue and communicating the outcomes.
- Raise the performance of learned control policies on internal evaluation suites and build on earlier policy work.
- Train models on large, uneven-quality collections of data where feedback is scarce, applying several value estimation approaches.
- Investigate model-based methods and learned environment models for continuous control.
- Examine what keeps reinforcement learning training stable as data, compute and model scale increase.
- Design rigorous experiments with dependable baselines and watch findings get tested on physical robots.
What we're looking for
- Background in reinforcement learning research, in a research or applied scientist capacity, connected to generative visual models.
- A record of papers accepted at leading machine learning or robotics venues.
- Depth in core and offline reinforcement learning topics, such as value estimation, goal-directed learning, policy design and training stability at scale.
- Familiarity with model-based methods for continuous control or with multimodal value estimation.
- An advanced degree (PhD or master's) in machine learning, computer science or an adjacent area, centered on reinforcement learning.
- Ability to work on-site six days a week.
Bonus points
- Prior exposure to visual generative models applied to robotics.
- Time spent doing research at a top-tier AI organization.
- Contributions to publicly shared reinforcement learning codebases.
- Experience with large-scale simulation, sim-to-real transfer or robotics, including JAX.
Compensation and benefits
- Base salary: USD 200,000 to 350,000 per year
- Equity
- Health, dental and vision coverage
Location and work model
- San Francisco, CA, United States
- On-site, 6 days per week in office
- Full-time