Research Scientist, Reinforcement Learning

About the role

You will own the design and execution of foundational research agendas that probe the mathematical principles underlying reasoning, learning, and decision-making within intelligent systems. This role requires you to construct and validate software implementations that embody these principles, ensuring that theoretical insights translate into robust and scalable behaviors. You will lead exploratory investigations into model-based methods, focusing on how agents construct, refine, and utilize world models through interactive experimentation. A core part of your ownership involves tackling concrete challenges such as the AutumnBench platform, physical and simulated robotics benchmarks, and the Abstract Reasoning Corpus (ARC) to assess generality. You will collaborate closely with interdisciplinary teams to define the research direction of the MARA project, shaping its trajectory and depth. Your work will emphasize rigorous, high-quality science where you are not afraid to tinker, iterate, and explore radically different ideas to solve hard problems. You will also translate complex findings into clear narratives that advance the broader field and influence subsequent research cycles.

Key facts

What you'll do

Develop and implement novel reinforcement learning and planning algorithms that advance the state of the art in model-based control and exploration.

Design and conduct experiments within physical and simulated robotics environments to evaluate the sample efficiency and generalization of learned policies.

Construct structured world models that enable agents to learn efficient policies in complex, partially observable environments through interactive experimentation.

Apply Bayesian optimization and optimal control techniques to automate decision-making processes and improve system performance over time.

Lead the integration of neural and symbolic methods to create AI systems that combine learning with explicit reasoning and abstraction.

Drive progress on the AutumnBench benchmark and other concrete challenges, ensuring that research remains grounded in measurable, real-world criteria.

Publish and present high-quality research findings at leading academic conferences and in peer-reviewed journals to disseminate knowledge to the broader community.

Mentor junior team members by providing technical guidance, fostering scientific rigor, and encouraging collaborative problem-solving across the team.

Contribute to the architectural direction of the MARA project by identifying open problems and defining research milestones that align with long-term goals.

Maintain and release open-source software artifacts that support reproducibility and enable external researchers to build upon your work effectively.

Participate in collaborative projects with external partners, aligning your expertise with broader efforts to solve intractable problems at scale.

Refine research methodologies to ensure that solutions are not only effective but also interpretable, generalizable, and aligned with foundational principles.

Champion a culture of experimentation where hypotheses are rigorously tested, mistakes are analyzed, and insights are systematically incorporated into future work.

Requirements

Researchers holding a PhD in computer science, artificial intelligence, machine learning, cognitive science, or related fields.

Strong background in reinforcement learning, planning, MDPs, optimal control, and sequential decision making.

Experience in developing AI systems that combine neural and symbolic methods is highly valued.

Interest in foundational AI research and its applications to modeling, abstraction, and reasoning.

Individuals with a demonstrated track record in scientific research, evidenced through publications, technical reports, or impactful software projects.

Excited about solving real world problems and having positive societal impact.

Eligible to work in the United States without sponsorship now or within 30 days of hire.

Authorized to work in the location listed on the application at the time of hire.

Willing and able to comply with Basis policies and applicable laws, rules, and regulations.

Must be able to commit to the essential functions of the role with or without reasonable accommodation.

Must be at least 18 years of age at the time of hire.

Must provide all required documentation evidence of identity and employment authorization as required by law.

Practical notes

In-person Policy: We are in the office four days a week. Be prepared to attend multi-day Basis-wide in-person events.

Location: This role is in-person in either New York City or Cambridge, MA.

FT/PT: This is a full-time position

Start date: Immediate start possible.

Salary range: Competitive salary.