Research Scientist Hierarchical RL Agents
A 501(c)(3) science-and-technology research institute/foundation operating direct Neuro-AI research, life-sciences programs through Radial, and an open-science residency program.
About Astera Institute
Astera Institute supports and operates open public-goods work for science and technology. Its current activities include neuroscience-informed AGI research, AI-enabled life-sciences programs, open datasets and publishing infrastructure, and residencies for early-stage projects.
Skills
Candidate Availability
Required and preferred rules are kept separate and reflect the wording in the original posting.
About the Role
You will develop and implement architectures and learning and inference algorithms for hierarchical reinforcement learning, world modeling, and planning. You will design ethologically constrained evaluation tasks, benchmark agent designs, and conduct experiments on cognition and learning. You will translate mathematical frameworks into scalable systems and contribute to open science.
Requirements
- Strong foundations in deep learning, graphical models, and information theory
- Expert-level proficiency with PyTorch or JAX
- Clean, reproducible research coding skills
- Demonstrated interest in AI robustness, generalization, and common-sense reasoning
- Passion for computational and architectural principles of the mammalian brain
- Ability to thrive in a lean, fast-paced environment with high autonomy
Responsibilities
- Develop and implement architectures and learning and inference algorithms for hierarchical reinforcement learning, world modeling, and planning
- Design ethologically constrained evaluation tasks and benchmark agent designs
- Design experiments on high-level cognition, causal representations, sensory-motor integration, active inference, and unsupervised learning
- Translate abstract mathematical frameworks into scalable systems with neuroscientists and software engineers
- Innovate publication models and support open-source code releases and open access