39.6 kB
Ctrl+K
- claim-1-for-learned-lagrangian-relaxation-multipliers-over-s-coupling-constraints-and-n-training-samples-the-expected-excess-risk-of-the-erm-solution-is-upper-bounded-by-o-s-1-5-sqrt-n-theorem-5-5
- claim-2-a-minimax-lower-bound-of-omega-s-sqrt-n-is-proven-showing-linear-dependence-on-the-number-of-coupled-constraints-s-is-unavoidable-for-any-learning-algorithm-theorem-5-6
- claim-3-stochastic-gradient-ascent-with-iterate-averaging-achieves-the-matching-o-s-sqrt-n-rate-and-is-minimax-optimal-theorem-5-12
- claim-4-learning-to-warm-start-subgradient-ascent-has-an-o-s-n-risk-upper-bound-and-matching-omega-s-n-minimax-lower-bound-theorems-6-1-and-6-2
- claim-5-the-multiplier-class-covering-number-and-its-entropy-integral-yield-rademacher-complexity-o-s-1-5-sqrt-n-lemmas-5-3-and-5-4
- claim-6-the-dual-function-is-concave-in-the-multiplier-and-has-subgradient-norm-at-most-2b-sqrt-s-under-bounded-violation-assumptions-proposition-5-1
- conclusion
- executive-summary
- 2.16 kB