First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper studies the multi-armed bandit problem where they have a set of relevant features; and the expected reward of an action is a Lipschitz continuous of relevant features. This is also a feature selection problem where you have a set of features but only r of them are relevant (the target function only depends on r of these features): here each arm has only one relevant feature, meaning the function representing the arm payoff depending on only one feature and we do not know which one. They propose an algorithm and get the bound for such adaptive case; but their regret is higher than what you would get if someone tells you the relevant type. Q2: Please summarize your review in 1-2 sentences This paper makes a small step towards understanding the problem of having a subset of features being relevant for a given arm which itself is certainly an interesting problem: they study the bandit problem only for one relevant feature per arm and did not give the optimal rate. Potentially, they could go with all arbitrary number of relevant features and figure out the optimal regret.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper investigate fast convergence properties of proximal gradient method and proximal Newton method under the assumption of Constant Nullspace Strong Convexity (CNSC). The problem of interest is to minimize the sum of two convex functions f(x)+h(x), where f is twice differentiable (smooth) and h can be non-smooth but admits a simple proximal mapping. Under the CNSC assumption on f and assuming h has the form of decomposable norm, this paper showed global geometric convergence of the proximal gradient method, and local quadratic convergence of the proximal Newton method. Writing of this paper is very clear.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper describes a non-negative matrix factorization for tall and skinny matrices. This algorithm works in the bigdata scenario because it need only pass over the tall skinny matrix one time. This linear read of the tall skinny matrix does not fully utilize the distributed mapreduce framework. I wonder, is it possible to parallelize the reading of the matrix and combine the results from subsets of the data into one final result?

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper studies the rank aggregation problem where a global ranking is inferred from multiple partial rankings. While assuming the partial rankings are generated according to the Plackett-Luce (PL) model, some of the results in the paper apply to the more general Thurstone's model as well. It provides theoretical results quantifying the required number of item assignments from users and analyzes the case where only pairwise comparisons are used as aggregation input. I find the results of the latter, i.e., rank-breaking upper bounds, especially interesting.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The authors propose a novel approach for hierarchical clustering of multivariate data. They construct cluster trees by estimating minimum volume sets using the q-One-Class SVM, and evaluate their method on a synthetic data set and two real word applications. While their new method seems to perform better than other approaches based on density estimation, I am not convinced by the benefits in practical applicability as the authors did not compare their method to the most commonly used hierarchical clustering techniques (agglomerative clustering with average linkage/ward). Minor comment: Rather than splitting their data once in a training and test set, the authors should perform 10-fold/5-fold cross-validation for a more reliable estimation of the generalizability of their method.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper introduces a system that creates a classifier at test time for a given test image using a set of training examples in a specific neighborhood of the test image. The authors method proposes how to identify the neighborhood with the most informative training examples for the given test image. The authors indicate that their method is well suited to fine-grained image classification tasks. The authors' key observation is that while training images that are nearest to the test image may be informative, the most informative set of training images must be found by estimating the informativeness of all members of a neighborhood set together. The main contribution of this paper is a method for estimating the most informative neighborhood set in an online fashion for each new test image.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper presents a spectral clustering algorithm for the sparse degree regime (degree=\theta(1)) of the stochastic blockmodel. The authors propose an alternative to the non-backtracking operator and point out that this has similar properties as the non-backtracking operator, which has been shown to be useful in the sparse regime. But the proposed data matrix is smaller than the non backtracking operator, and symmetric, therefore making eigenvalue computation easier and more accurate. The sparse regime is indeed the hardest in terms of showing concentration of empirical eigenvalues, or performance of a clustering algorithm in the stochastic blockmodel.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper defines temporal coverage functions, a generalization of coverage functions with dependence. It introduces a method to learn them from previous time history. It also includes experimental validation on real data sets. The paper is well written, and clear. It is also very technical, and was not an easy read for me.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper proposes an algorithm for online combinatorial optimization. In this online learning problem, the action space is combinatorially large and can be represented in a d-dimensional Euclidean space such that the loss in each time step is a linear function of the action. It would greatly improve the paper if there was a thorough comparison between the new algorithm and Online Stochastic Mirror Descent (OSMD by Audibert et al., [3] in the current paper) both in terms of how the algorithms work and in terms of regret bounds. In the current form of the paper, I am not sure if the new algorithm is significantly different from OSMD or if it improves its bounds.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper addresses the problem of learning with label proportions, wherein one only knows the proportions of the labels of bags of samples that are positive (or more generally that belong to one class). It makes very substantive contributions by: a. analysis that shows that a general class of loss functions allow efficient learning via the mean operator without requiring homogeneity (as was the case with previous literature) b. Developing fast learning algorithms for estimating the mean operator c. allow the use of standard binary classifier learning algorithms to solve the LLP problem via reduction. It is also very well written, though it is quite dense because of the sheer volume of novel material that the authors try to cover in a short NIPS paper.