"NIPS Neural Information Processing Systems 8-11th December 2014, Montreal, Canada",,, "Paper ID:","1612" "Title:","Improved Distributed Principal Component Analysis" Current Reviews First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper considers the problem of trading off the communication and computation cost of distributed computation and proposes a new distributed k L-2 error fitting algorithm. The proposed algorithm can be seen as a combination of many previous speed up techniques for distributed PCA and clustering methods. However, the authors also contribute optimizations over the base methods and further improves the communication and computation efficiency. The theoretical guarantee is sound and experiments are convincing.

Neural Information Processing Systems http://nips.cc/

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The paper develops lower and upper bounds on the required rank of adjacency tensor factorizations to recover the adjacency tensor. This is, it investigates the problem of the minimal rank required to express the true underlying data via factorizations. These bounds are shown to of practical use by scaling RESCAL up. The paper is extremely well written and makes several interesting and important contributions.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The paper studies the interesting question on online (stochastic) gradient descent in the unconstrained setting (sometime referred to non-strongly convexity or without explicit regularization). In the earlier work [33], it was proved that the last iterate of stochastic gradient descent with least-square loss in the unconstrained setting actually converges with explicit convergence rate by appropriately choosing the step sizes (or by a stopping early rule), which, however, needs to know the smoothness of the regression function. The paper proposed a kernel-based stochastic gradient descent algorithm without the need to perform model selection, which requires the loss function and its gradient are both Lipschitz. The proposed algorithm is mainly motivated by the recent studies [15,16] which involves a data-dependent regularization.

Neural Information Processing Systems http://nips.cc/

Neural Information Processing Systems http://nips.cc/