Proposes a nonparametric RKHS embedding approach for modeling MDP transition dynamics, outperforming Gaussian processes and NPDP in efficiency and accuracy.

Steffen Grunewalder, Guy Lever, Luca Baldassarre et al.

Proposes a training-data-driven expert learning algorithm with context trees for improved short-sequence prediction.

Elad Eban, Aharon Birnbaum, Shai Shalev-Shwartz et al.

Revealed the equivalence between conditional mean embeddings and vector-valued regression, achieving near-optimal convergence rate of O(log(n)/n) with sparse regularization.

Steffen Grünewälder, Guy Lever, Luca Baldassarre et al.

This paper rigorously analyzes regret bounds for stochastic and adversarial multi-armed bandits, establishing near-optimal guarantees for UCB and Exp3 algorithms.

Sébastien Bubeck, Nicolò Cesa-Bianchi

Scikit-learn is a Python library integrating state-of-the-art ML algorithms like SVM and PCA, emphasizing usability and performance.

Fabian Pedregosa, Gaël Varoquaux, Alexandre Gramfort et al.

Large-scale unsupervised deep autoencoder trained on 10 million images achieves high-level feature detection, boosting ImageNet accuracy by 70%.

Quoc V. Le, Marc'Aurelio Ranzato, Rajat Monga et al.

Analysis of 0-1 loss reveals superior noise tolerance, especially under non-uniform label noise, with theoretical guarantees and empirical validation.

Naresh Manwani, P. S. Sastry

Proposes a model-free, weighted risk-sensitive RL algorithm balancing reward and error state probability, validated on continuous and discrete control tasks.

P. Geibel, F. Wysotzki

Proposes Gossip learning with linear models using random walks and model averaging, enabling privacy-preserving, low-communication distributed classification.

Róbert Ormándi, István Hegedüs, Márk Jelasity

Proposes a unified neural network framework leveraging large-scale unlabeled data for multiple NLP tasks, achieving state-of-the-art results without task-specific engineering.

Ronan Collobert, Jason Weston, Leon Bottou et al.

Bayesian optimization efficiently maximizes expensive black-box functions using Gaussian processes and acquisition functions, reducing sample complexity.

Eric Brochu, Vlad M. Cora, Nando de Freitas

Introduces LinUCB, a linear contextual Bandit algorithm, achieving 12.5% click rate lift on Yahoo! dataset, suitable for large-scale personalized news recommendation.

Lihong Li, Wei Chu, John Langford et al.

Exp4.P algorithm achieves supervised learning-like guarantees in contextual bandit problems, significantly reducing regret.

Alina Beygelzimer, John Langford, Lihong Li et al.

Proposes a spectral matrix completion algorithm achieving O(rn) sample efficiency with provable error bounds.

Raghunandan H. Keshavan, Andrea Montanari, Sewoong Oh

Linearly Parameterized Bandits use exploration-exploitation strategy to achieve Θ(r√T) cumulative regret and Bayes risk.

Paat Rusmevichientong, John N. Tsitsiklis

Proposes a kernel-based two-sample test framework (MMD) for efficient distribution comparison with strong theoretical guarantees.

Arthur Gretton, Karsten Borgwardt, Malte J. Rasch et al.

Conformal prediction provides confidence sets with guaranteed coverage probability in online settings, applicable to models like SVM and ridge regression.

Glenn Shafer, Vladimir Vovk