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

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

We focus on the case of i.i.d.

Humans can quickly learn new tasks from just a handful of examples by preserving rich representations of experience [1].

We give a tractable unbiased estimate of the log density using a "Russian roulette" estimator, and reduce the memory required

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. The paper presents new techniques, and theoretical results relating to those techniques, to optimise the Variational Gaussian lower bound to the log evidence in latent Gaussian models LGMs. The technique could be applied to a broader category of latent linear models but their presentation focuses on LGMs only. VG approximate inference is important because it has many nice properties: it is widely applicable, often quite accurate and relatively fast. This paper attempts to makes VG methods more scalable without resorting to making factorisation assumptions on the approximating Gaussian distribution. Whilst the authors show how the objective function can be `decoupled' they do not show experimentally that this leads to a clear improvement in speed or scalability over standard techniques.

Bayesian decision theory outlines a rigorous framework for making optimal decisions based on maximizing expected utility over a model posterior. However, practitioners often do not have access to the full posterior and resort to approximate inference strategies.

They are themselves constructed from i.i.d. Gaussian processes, in analogy to the construction of Wishart or inverse Wishart random variables from i.i.d.

Quasi-Monte Carlo (QMC) is an essential tool for integral approximation, Bayesian inference, and sampling for simulation in science, etc.