"NIPS Neural Information Processing Systems 8-11th December 2014, Montreal, Canada",,, "Paper ID:","1233" "Title:","A Multiplicative Model for Learning Distributed Text-Based Attribute Representations" Current Reviews First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper proposes to incorporate side information for improving vector-space embedding of words via an attribute vector that modulates the word-projection matrices. One could simply think of word-projection tensors (although, in practice the tensors are factorized) where the attribute vector provide the loadings for the tensor slices. This is studied in the context of log-bilinear language models, but the basic idea should be applicable to other word embedding work. The theory part of the paper is very well-written. However, it is in the experimental section that things get somewhat muddier.

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. This paper proposes a new regularization method for structured prediction. The idea is relatively straightforward: a linear chain model is segmented into smaller subchains, each of which is added as an independent training example. Theorems are provided (with proofs in the supplement) showing how this regularization can reduce generalization risk and accelerate convergence rates. Empirical comparisons with state of the art approaches suggest that the resulting method is both faster and more accurate.

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

We identify the marginal polytope, the output space's convex hull,

"NIPS Neural Information Processing Systems 8-11th December 2014, Montreal, Canada",,, "Paper ID:","157" "Title:","Object Localization based on Structural SVM using Privileged Information" Current Reviews First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The method is effective for the object localization task and results in good improvements in localization accuracy. It looks like the authors' formulation of SSVM+ contains separate slack variables \xi_i for each example x_i and there are extra degrees of freedom. How many alternating iterations are required? When the parameter vectors w and w^* are far from the optimal solution, could this alternating inference procedure get stuck in bad local minima?

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

Structured prediction can be thought of as a simultaneous prediction of multiple labels.

Modern prediction problems arising in multilabel learning and learning to rank pose unique challenges to the classical theory of supervised learning. These problems have large prediction and label spaces of a combinatorial nature and involve sophisticated loss functions. We offer a general framework to derive mistake driven online algorithms and associated loss bounds. The key ingredients in our framework are a general loss function, a general vector space representation of predictions, and a notion of margin with respect to a general norm. Our general algorithm, Predtron, yields the perceptron algorithm and its variants when instan-tiated on classic problems such as binary classification, multiclass classification, ordinal regression, and multilabel classification. For multilabel ranking and subset ranking, we derive novel algorithms, notions of margins, and loss bounds. A simulation study confirms the behavior predicted by our bounds and demonstrates the flexibility of the design choices in our framework.

In user-facing applications, displaying calibrated confidence measures-- probabilities that correspond to true frequency--can be as important as obtaining high accuracy. We are interested in calibration for structured prediction problems such as speech recognition, optical character recognition, and medical diagnosis. Structured prediction presents new challenges for calibration: the output space is large, and users may issue many types of probability queries (e.g., marginals) on the structured output. We extend the notion of calibration so as to handle various subtleties pertaining to the structured setting, and then provide a simple recalibra-tion method that trains a binary classifier to predict probabilities of interest. We explore a range of features appropriate for structured recalibration, and demonstrate their efficacy on three real-world datasets.