Sohrab Andaz
I am a Senior Research Scientist at Amazon, where I work on decision-making under uncertainty using reinforcement learning. My work is typically focused on optimizing real, large-scale supply chains, but I’m broadly interested in reinforcement learning—both single-agent and multi-agent—for large-scale systems.
Research
My work has primarily focused on inventing and deploying novel algorithms for supply chain systems. Traditional methods in reinforcement learning fail to account for important facets of the real-world. My work aims to fill gaps around safe exploration, decision-making with operational constraints, sim-to-real transfer, and integration with traditional operations research solvers and software stacks, with the goal of deploying solutions to production. Beyond real-world reinforcement learning, I'm also generally interested in simulator construction (specifically differentiability), neural approximators for reinforcement learning, meta-learning, and open-ended learning.
Contact
The fastest way to reach me is s.s.andaz@gmail.com.