Thank you for all the helpful comments. Several related works were raised by the reviewers which we discuss here. We note that the authors have marked their ArXiv submission as containing errors. Each of their inner loops uses SGD to solve the distance-regularized objectives. First, we use the EMA of slow weights to adjust the training parameters during optimization.

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

We study the secure stochastic convex optimization problem.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The paper introduces a novel convex region-specific linear models called partition-wise linear model. It assigns linear models to partitions of the input space and linear combination of these partition-specific models define the region-specific linear models. This allows them to construct convex objective functions. They optimize both the regions and predictors by using sparsity inducing structured penalties.

Unsupervised methods have been recently used to produce more compact hashes than their randomized equivalents. The authors demonstrate that existing methods suffer from bad performance as the length of the codes increases and suggest a new graph-based method. To achieve better codes, they keep the binary constraints and consider a slightly relaxed formulation (still NP-hard) that they solve using alternating maximization. Local convergence is guaranteed and extensive experiments show that the suggested method achieves very good performance on a number of datasets.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. Summary: This paper attempts to link sparse optimization methodology to the anatomical structure of locust's early olfactory system. The work is motivated by the observation that odorant molecules are sparsely represented by the population of Kenyon cells. The authors first mathematically formulate the olfactory system as a MAP decoder, and give the standard solution to the problem without considering biological constraints. Next, to make the solution more biologically plausible, the authors reformulate the olfactory system model as a decoder of a compressive sensing problem, and provide two standard solutions to the dual problem. Then, the authors argue that each of the components in the solution can be mapped/interpreted to/as a unit of the biological structure in the olfactory system. However, these maps are described without a strong justification and there are conceptual problems in linking the math with the biology.

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

Therefore, when used as a per-processing step in sparse group lasso problem to reduce its scale, the computational complexity of the optimization problem can be reduced significantly without sacrificing the accuracy of the model.