The reward is part of the dataset
Three measurements that matter for kernel RL: correct speedups, trustworthy rewards, and learning that transfers. A research agenda for evaluating the data behind kernel-writing agents.
THE NODUS RESEARCH JOURNAL
Methods, protocols, and results for GPU kernel learning. Follow the sources, understand the measurements, and put the findings to work.
Follow new publicationsThree measurements that matter for kernel RL: correct speedups, trustworthy rewards, and learning that transfers. A research agenda for evaluating the data behind kernel-writing agents.