Statistical Learning
From Biased to Unbiased Dynamics: An Infinitesimal Generator Approach
A general idea to overcome the above problem is to perturb the system dynamics.
RL: Efficient Exploration for Nonepisodic RL
RL achieves the optimal average cost while incurring the least regret.
The Implicit Bias of Gradient Descent toward Collaboration between Layers: A Dynamic Analysis of Multilayer Perceptions
Another intriguing phenomenon is the existence of adversarial examples -- imperceptible perturbations of inputs that alter classification results.
Gaussian Process Bandits for Top-k Recommendations
Theoretical analysis confirms that the proposed algorithm achieves sub-linear regret in relation to the number of rounds and arms.
Oja's Algorithm for Streaming Sparse PCA
This is completely different from previous analyses of Oja's algorithm and matrix products, which
Small steps no more: Global convergence of stochastic gradient bandits for arbitrary learning rates
In particular, we establish the surprising result that: F or any constant learning rate η > 0, the stochastic gradient bandit algorithm is guaranteed to converge to the globally optimal policy almost surely.
Unity by Diversity: Improved Representation Learning for Multimodal V AEs Thomas M. Sutter
Such architectures impose hard constraints on the model.