Replicable algorithms for heterogeneous problems can be composed with O(sum n_i) samples at constant replicability via conversion to perfectly generalizing algorithms, privacy-style composition, and correlated sampling.
Private selection from private candidates
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Replicable Composition
Replicable algorithms for heterogeneous problems can be composed with O(sum n_i) samples at constant replicability via conversion to perfectly generalizing algorithms, privacy-style composition, and correlated sampling.