Though with progress, model learning and performing posterior inference still remains a common challenge for using deep generative models, especially for handling discrete hidden variables. This paper is mainly concerned with algorithms for learning Helmholz machines, which is characterized by pairing the generative model with an auxiliary inference model. A common drawback of previous learning algorithms is that they indirectly optimize some bounds of the targeted marginal log-likelihood. In contrast, we successfully develop a new class of algorithms, based on stochastic approximation (SA) theory of the Robbins-Monro type, to directly optimize the marginal log-likelihood and simultaneously minimize the inclusive KL-divergence. The resulting learning algorithm is thus called joint SA (JSA). Moreover, we construct an effective MCMC operator for JSA. Our results on the MNIST datasets demonstrate that the JSA's performance is consistently superior to that of competing algorithms like RWS, for learning a range of difficult models.

We analyze a fast incremental aggregated gradient method for optimizing nonconvex problems of the form $\min_x \sum_i f_i(x)$. Specifically, we analyze the SAGA algorithm within an Incremental First-order Oracle framework, and show that it converges to a stationary point provably faster than both gradient descent and stochastic gradient descent. We also discuss a Polyak's special class of nonconvex problems for which SAGA converges at a linear rate to the global optimum. Finally, we analyze the practically valuable regularized and minibatch variants of SAGA. To our knowledge, this paper presents the first analysis of fast convergence for an incremental aggregated gradient method for nonconvex problems.

Bayesian networks are probabilistic graphical models that represent a set of random variables and their conditional dependencies via a directed acyclic graph. Explicitly modeling the conditional dependencies between random variables permit efficient algorithms to perform inference and learning in the network. Causal Bayesian networks have the additional requirement that all edges in the network model a causal relationship. Dynamic Bayesian networks are the time-generalization of Bayesian networks and relate variables to each other over adjacent time steps. Dynamic Bayesian networks unify and extend a number of state-space models including hidden Markov models, hierarchical hidden Markov models and Kalman filters. Dynamic Bayesian networks can also be seen as the natural extension of acyclic causal models to models that permit cyclic causal relationships, while avoiding problems with causal models that try to model temporal relationships with an atemporal description [1]. A natural question is what is the expressive power of such networks. The result in this paper shows that although discrete dynamic Bayesian networks are sub-Turing in computational power, introducing continuous random variables with discrete children is sufficient to model Turing-complete computation.

We consider the problem of packing node-disjoint directed paths in a directed graph. We consider a variant of this problem where each path starts within a fixed subset of root nodes, subject to a given bound on the length of paths. This problem is motivated by the so-called kidney exchange problem, but has potential other applications and is interesting in its own right. We propose a new algorithm for this problem based on the message passing/belief propagation technique. A priori this problem does not have an associated graphical model, so in order to apply a belief propagation algorithm we provide a novel representation of the problem as a graphical model. Standard belief propagation on this model has poor scaling behavior, so we provide an efficient implementation that significantly decreases the complexity. We provide numerical results comparing the performance of our algorithm on both artificially created graphs and real world networks to several alternative algorithms, including algorithms based on integer programming (IP) techniques. These comparisons show that our algorithm scales better to large instances than IP-based algorithms and often finds better solutions than a simple algorithm that greedily selects the longest path from each root node. In some cases it also finds better solutions than the ones found by IP-based algorithms even when the latter are allowed to run significantly longer than our algorithm.

Learning the "blocking" structure is a central challenge for high dimensional data (e.g., gene expression data). Recently, a sparse singular value decomposition (SVD) has been used as a biclustering tool to achieve this goal. However, this model ignores the structural information between variables (e.g., gene interaction graph). Although typical graph-regularized norm can incorporate such prior graph information to get accurate discovery and better interpretability, it fails to consider the opposite effect of variables with different signs. Motivated by the development of sparse coding and graph-regularized norm, we propose a novel sparse graph-regularized SVD as a powerful biclustering tool for analyzing high-dimensional data. The key of this method is to impose two penalties including a novel graph-regularized norm ($|\pmb{u}|\pmb{L}|\pmb{u}|$) and $L_0$-norm ($\|\pmb{u}\|_0$) on singular vectors to induce structural sparsity and enhance interpretability. We design an efficient Alternating Iterative Sparse Projection (AISP) algorithm to solve it. Finally, we apply our method and related ones to simulated and real data to show its efficiency in capturing natural blocking structures.

We study large-scale kernel methods for acoustic modeling and compare to DNNs on performance metrics related to both acoustic modeling and recognition. Measuring perplexity and frame-level classification accuracy, kernel-based acoustic models are as effective as their DNN counterparts. However, on token-error-rates DNN models can be significantly better. We have discovered that this might be attributed to DNN's unique strength in reducing both the perplexity and the entropy of the predicted posterior probabilities. Motivated by our findings, we propose a new technique, entropy regularized perplexity, for model selection. This technique can noticeably improve the recognition performance of both types of models, and reduces the gap between them. While effective on Broadcast News, this technique could be also applicable to other tasks.

Fault detection in industrial plants is a hot research area as more and more sensor data are being collected throughout the industrial process. Automatic data-driven approaches are widely needed and seen as a promising area of investment. This paper proposes an effective machine learning algorithm to predict industrial plant faults based on classification methods such as penalized logistic regression, random forest and gradient boosted tree. A fault's start time and end time are predicted sequentially in two steps by formulating the original prediction problems as classification problems. The algorithms described in this paper won first place in the Prognostics and Health Management Society 2015 Data Challenge.

This paper studies the Best-of-K Bandit game: At each time the player chooses a subset S among all N-choose-K possible options and observes reward max(X(i) : i in S) where X is a random vector drawn from a joint distribution. The objective is to identify the subset that achieves the highest expected reward with high probability using as few queries as possible. We present distribution-dependent lower bounds based on a particular construction which force a learner to consider all N-choose-K subsets, and match naive extensions of known upper bounds in the bandit setting obtained by treating each subset as a separate arm. Nevertheless, we present evidence that exhaustive search may be avoided for certain, favorable distributions because the influence of high-order order correlations may be dominated by lower order statistics. Finally, we present an algorithm and analysis for independent arms, which mitigates the surprising non-trivial information occlusion that occurs due to only observing the max in the subset. This may inform strategies for more general dependent measures, and we complement these result with independent-arm lower bounds.

A thesis submitted for the degree of Doctor of Philosophy of The Australian National University. In this work we introduce several new optimisation methods for problems in machine learning. Our algorithms broadly fall into two categories: optimisation of finite sums and of graph structured objectives. The finite sum problem is simply the minimisation of objective functions that are naturally expressed as a summation over a large number of terms, where each term has a similar or identical weight. Such objectives most often appear in machine learning in the empirical risk minimisation framework in the non-online learning setting. The second category, that of graph structured objectives, consists of objectives that result from applying maximum likelihood to Markov random field models. Unlike the finite sum case, all the non-linearity is contained within a partition function term, which does not readily decompose into a summation. For the finite sum problem, we introduce the Finito and SAGA algorithms, as well as variants of each. For graph-structured problems, we take three complementary approaches. We look at learning the parameters for a fixed structure, learning the structure independently, and learning both simultaneously. Specifically, for the combined approach, we introduce a new method for encouraging graph structures with the "scale-free" property. For the structure learning problem, we establish SHORTCUT, a O(n^{2.5}) expected time approximate structure learning method for Gaussian graphical models. For problems where the structure is known but the parameters unknown, we introduce an approximate maximum likelihood learning algorithm that is capable of learning a useful subclass of Gaussian graphical models.

This paper offers a characterization of fundamental limits on the classification and reconstruction of high-dimensional signals from low-dimensional features, in the presence of side information. We consider a scenario where a decoder has access both to linear features of the signal of interest and to linear features of the side information signal; while the side information may be in a compressed form, the objective is recovery or classification of the primary signal, not the side information. The signal of interest and the side information are each assumed to have (distinct) latent discrete labels; conditioned on these two labels, the signal of interest and side information are drawn from a multivariate Gaussian distribution. With joint probabilities on the latent labels, the overall signal-(side information) representation is defined by a Gaussian mixture model. We then provide sharp sufficient and/or necessary conditions for these quantities to approach zero when the covariance matrices of the Gaussians are nearly low-rank. These conditions, which are reminiscent of the well-known Slepian-Wolf and Wyner-Ziv conditions, are a function of the number of linear features extracted from the signal of interest, the number of linear features extracted from the side information signal, and the geometry of these signals and their interplay. Moreover, on assuming that the signal of interest and the side information obey such an approximately low-rank model, we derive expansions of the reconstruction error as a function of the deviation from an exactly low-rank model; such expansions also allow identification of operational regimes where the impact of side information on signal reconstruction is most relevant. Our framework, which offers a principled mechanism to integrate side information in high-dimensional data problems, is also tested in the context of imaging applications.