To solve this, we perform probabilistic reasoning over the depth of neural networks.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. COMMENTS BASED ON REVIEWER DISCUSSIONS AND AUTHOR REBUTTAL: I agree with the other reviewers that more could be done to constrain the specifics of the cue integration mechanism. However, I believe that if the data set is expanded, allowing the models to be better constrained, then the paper is appropriate and interesting for the NIPS community. I have left my quality score as it was, but I agree with the other reviewers that the paper merits a ``1'' rather than a ``2'' for impact score. ORIGINAL REVIEW: Summary: This paper extends an existing model for the perception of visual speed that uses a Bayesian observer model acting on the activity of independent spatiotemporal frequency channels. Previously, the model accounted for illusions of perceived speed by postulating the Bayes-optimal combination of noisy sensory representations with a prior for slow speeds.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper addresses the challenge of inferring synaptic connectivity in settings where one or more putative presynaptic neurons are stimulated while the membrane potential of a single postsynaptic neuron is recorded, as in modern two-photon microscopy experiments. The authors present a new probabilistic model for this experimental setup, along with a variational inference algorithm. They then develop an online, active learning algorithm to select the most informative stimuli. The efficacy of their algorithm is demonstrated using synthetic data generated from the model.

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. UPDATE: I acknowledge that I have read the author rebuttal. The authors propose a method for learning a mapping from input messages to the output message in the context of expectation propagation. The method can be thought of as a sort of compilation step, where there is a one-time cost of closely approximating the true output messages using important sampling, after which a neural network is trained to reproduce the output messages in the context of future inference queries. First, the authors should be commended for attacking a difficult and interesting problem.

The paper introduces a probabilistic model for networks which assigns each node in the network to multiple, overlapping latent communities. Inference is done using a stochastic variational method and the experimental evaluations are performed on very large networks. The first thing I note is that you do not cite Morup et al. (2010) "Infinite multiple membership relational modelling for complex networks", which in truth was the first work to perform inference for a latent feature relational model on large datasets -- in effect, rendering your statement on 067-068 "... these innovations allow the first..." incorrect. This is a rather serious oversight, because their paper not only performs large scale inference, but their method is also an MCMC method, which is well-known to usually produce more accurate results than variational methods. I believe the strongest contribution from this paper is the application of a stochastic variational inference method to a relational data model.

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This is an interesting paper - the application of graphical methods to analyze missing data patterns may prove to be very useful. The paper contains the word causal graph in the title and in the introduction. However, none of the results seem to depend on the graph being causal. The results are entirely about conditional independence and factorizations.