Mohamed Ayman Mohamed

I work on continual reinforcement learning, the setting where an agent never gets a final task, only a world that keeps moving.

I did my MSc at the University of Alberta with Marlos C. Machado. Most of my time there went into a problem that sounds like bookkeeping and turns out not to be: we had no good place to study agents that must keep adapting. Benchmarks hand you a fixed task and ask how well you eventually solve it. So I built one that refuses to hold still, then measured how the standard algorithms cope. They largely do not.

Alongside that I work on multi-agent cooperation, asking how agents can value each other's behaviour without ever seeing each other's rewards. Earlier, at KAUST, I worked on grounding language in 3D scenes, which is where the ICLR and TPAMI papers below come from.

Day to day I am a software engineer at AWS, building AI systems for Kubernetes and region infrastructure. Before research I spent a long time close to the metal: pipelined RISC-V in Verilog, an x86-64 bootloader that walks itself into long mode, a distributed system in Rust with its own election protocol. That background still shapes how I work: I would rather measure the quantity everyone assumes than assume it.