Ctrl+K
- claim-1-theorem-5-1-gives-first-convergence-bound-federated-dpo
- claim-2-corollary-5-2-full-participation-variance-amplification
- claim-3-theorem-5-4-introduces-staleness-penalty-term
- claim-4-theorem-5-5-establishes-lower-bound-showing-dependence
- claim-5-theorem-6-1-decdpo-converges-rate-where-spectral-gap
- claim-6-section-7-numerical-experiments-stanford-human
- conclusion
- executive-summary
- 1.01 kB