New paper proves fundamental confidence-efficiency bounds for transductive conformal prediction
_onionesque · x · 2026-09-23
An arXiv paper by Behboodi et al. establishes fundamental bounds for transductive conformal prediction: any non-trivial confidence level forces prediction sets to grow exponentially, with an exponent scaling linearly in sample count and proportional to the data's conditional entropy, plus a second-order dispersion term. The bound is achievable in an idealized setting, and the same-label special case reduces to hypothesis testing with empirically observed statistics, yielding an asymptotically optimal predictor.
More from Research
- Mixture-of-Depths routes FLOPs per token to cut transformer compute — burny_tech · 2026-09-23
- AI has solved the hardest part of formal verification, says Theorem co-founder — burny_tech · 2026-09-23
- Minerva framework preprint is out, says Brian Hie — BrianHie · 2026-09-23
- Minerva: Hie lab's data-mining framework for scientific discovery — BrianHie · 2026-09-23
- Genome language models uncover new class of reverse-transcriptase mechanisms — BrianHie · 2026-09-23
- Mathematician shares a cheap 4-step heuristic for hyperparameter tuning — dejanseo · 2026-09-23