September 2026
Elimination in Fusing Bandits: When Should We Stop Measuring?
From RAGE to witness-bottleneck elimination: how fusion changes which directions a bandit learner still needs to measure.
Technical writing, implementations, research notes, and materials that document both results and the reasoning behind them.
From RAGE to witness-bottleneck elimination: how fusion changes which directions a bandit learner still needs to measure.
Curriculum learning as an outer decision problem, from task-level bandits to problem-level selection in LLM post-training.
Why rewards and preferences are different measurements—and what it would take for both to inform one policy learner.
How an apparently successful RL experiment taught me to stop treating algorithm names as causal explanations.
How a Machine Learning Systems perspective changed a Bayesian routing project into a question about scale and resource allocation.
Study notes, implementations, and experiments covering foundational machine-learning topics.