Follow-the-Regularized-Leader
PulseAugur coverage of Follow-the-Regularized-Leader — every cluster mentioning Follow-the-Regularized-Leader across labs, papers, and developer communities, ranked by signal.
4 day(s) with sentiment data
-
New paper details minimax-optimal bounds for online linear optimization
A new paper published on arXiv details advancements in online linear optimization on $\ell_p$ balls. The research introduces a Follow-the-Regularized-Leader (FTRL) algorithm that achieves minimax-optimal regret bounds, …
-
New algorithms tackle prediction with expert advice under heavy-tailed losses
Researchers have developed new algorithms for prediction with expert advice, specifically addressing scenarios with heavy-tailed losses. Existing methods can be suboptimal when losses have a bounded moment less than one…
-
New algorithm achieves optimal regret for adversarial linear CMDPs
Researchers have developed a new primal-dual algorithm that achieves optimal $\widetilde{\mathcal{O}}(\sqrt{K})$ regret and cumulative constraint violation for episodic adversarial linear CMDPs with unknown transitions.…
-
New research tackles dynamic regret in online learning with novel methods · 3 sources tracked
Three new research papers explore dynamic regret minimization in online learning settings. The first paper focuses on Follow-the-Regularized-Leader (FTRL) methods for curved losses and adaptive optimizers like Adam, pre…
-
New research explores 1/2-Tsallis-INF algorithm for best-arm identification
This research paper investigates the effectiveness of the 1/2-Tsallis-INF algorithm, a known method for minimizing regret in multi-armed bandit problems, when applied to the task of identifying the best arm. The study a…
-
New algorithm achieves $\tilde{O}(T^{-1/4})$ convergence in bilinear saddle-point problems
Researchers have developed a new uncoupled learning algorithm that achieves last-iterate convergence to Nash equilibrium in bilinear saddle-point problems. This algorithm guarantees convergence with a rate of $\tilde{O}…
-
New adaptive learning rates enhance Follow-the-Perturbed-Leader algorithm
Researchers have developed a new adaptive learning rate method for the Follow-the-Perturbed-Leader (FTPL) algorithm in online learning. This approach introduces surrogate probability functions to enable probability-depe…
-
Paper analyzes strategic surplus against FTRL learners in games
This paper investigates the strategic surplus achievable against a Follow-the-Regularized-Leader (FTRL) learning algorithm in two-player zero-sum games. The research demonstrates that the extraction of this surplus is a…