Implemented PPI++ v2 adaptive power tuning, trading off Type I error and power
IanArawjo · x · 2026-08-16
Ian Arawjo implemented adaptive power tuning (PPI++ "v2") for all PPI methods. Compared to the previous run, the new version achieves similar average Type I control and statistical power, with only a very minor hit as expected by theory. The Wilcoxon method showed significantly lower power, and a fix is currently being tested.
More from Research
- New Study: Training a Single Transformer Layer Can Match Full-Parameter RL — tokenbender · 2026-08-16
- Study reveals 'latent programming horizon' allowing coding agents to predict future outcomes — tokenbender · 2026-08-16
- Developer Tinkers with Reinforcement Learning Late Night — sharpeye_wnl · 2026-08-16
- Open Source 10K Execution-Verified Financial Problems Dataset — coslinedev · 2026-08-16
- Can LLMs realize we're solving the problem wrong? Paradigm shifts in AI — ChippHop · 2026-08-16
- Physicist Uses Claude AI to Tackle Open Problem in Thermodynamics — bronzeagepapi · 2026-08-16