Follow-the-Perturbed-Leader with Pareto perturbations reaches the optimal O(sqrt(mdT)) regret in adversarial size-invariant combinatorial semi-bandits, and a conditional resampling variant cuts per-round complexity to O(md(log(d/m)+1)).
Regret in online combinatorial optimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Note on Follow-the-Perturbed-Leader in Combinatorial Semi-Bandit Problems
Follow-the-Perturbed-Leader with Pareto perturbations reaches the optimal O(sqrt(mdT)) regret in adversarial size-invariant combinatorial semi-bandits, and a conditional resampling variant cuts per-round complexity to O(md(log(d/m)+1)).