In reinforcement learning, temporal difference (TD) is the most direct algorithm to learn the value function of a policy. For large or infinite state spaces, exact representations of the value function are usually not available, and it must be approximated by a function in some parametric family. However, with \emph{nonlinear} parametric approximations (such as neural networks), TD is not guaranteed to converge to a good approximation of the true value function within the family, and is known to diverge even in relatively simple cases. TD lacks an interpretation as a stochastic gradient descent of an error between the true and approximate value functions, which would provide such guarantees. We prove that approximate TD is a gradient descent provided the current policy is \emph{reversible}. This holds even with nonlinear approximations. A policy with transition probabilities $P(s,s')$ between states is reversible if there exists a function $\mu$ over states such that $\frac{P(s,s')}{P(s',s)}=\frac{\mu(s')}{\mu(s)}$. In particular, every move can be undone with some probability. This condition is restrictive; it is satisfied, for instance, for a navigation problem in any unoriented graph. In this case, approximate TD is exactly a gradient descent of the \emph{Dirichlet norm}, the norm of the difference of \emph{gradients} between the true and approximate value functions. The Dirichlet norm also controls the bias of approximate policy gradient. These results hold even with no decay factor ($\gamma=1$) and do not rely on contractivity of the Bellman operator, thus proving stability of TD even with $\gamma=1$ for reversible policies.

In this work we present a modified neural network model which is capable to simulate Markov Chains. We show how to express and train such a network, how to ensure given statistical properties reflected in the training data and we demonstrate several applications where the network produces non-deterministic outcomes. One example is a random walker model, e.g.

We propose a Monte Carlo algorithm to sample from high-dimensional probability distributions that combines Markov chain Monte Carlo (MCMC) and importance sampling. We provide a careful theoretical analysis, including guarantees on robustness to high-dimensionality, explicit comparison with standard MCMC and illustrations of the potential improvements in efficiency. Simple and concrete intuition is provided for when the novel scheme is expected to outperform standard schemes. When applied to Bayesian Variable Selection problems, the novel algorithm is orders of magnitude more efficient than available alternative sampling schemes and allows to perform fast and reliable fully Bayesian inferences with tens of thousands regressors.

Dynamic topic modeling facilitates the identification of topical trends over time in temporal collections of unstructured documents. We introduce a novel unsupervised neural dynamic topic model named as Recurrent Neural Network-Replicated Softmax Model (RNNRSM), where the discovered topics at each time influence the topic discovery in the subsequent time steps. We account for the temporal ordering of documents by explicitly modeling a joint distribution of latent topical dependencies over time, using distributional estimators with temporal recurrent connections. Applying RNN-RSM to 19 years of articles on NLP research, we demonstrate that compared to state-of-the art topic models, RNNRSM shows better generalization, topic interpretation, evolution and trends. We also introduce a metric (named as SPAN) to quantify the capability of dynamic topic model to capture word evolution in topics over time.

A group of transition probability functions form a Shannon's channel whereas a group of truth functions form a semantic channel. Label learning is to let semantic channels match Shannon's channels and label selection is to let Shannon's channels match semantic channels. The Channel Matching (CM) algorithm is provided for multi-label classification. This algorithm adheres to maximum semantic information criterion which is compatible with maximum likelihood criterion and regularized least squares criterion. If samples are very large, we can directly convert Shannon's channels into semantic channels by the third kind of Bayes' theorem; otherwise, we can train truth functions with parameters by sampling distributions. A label may be a Boolean function of some atomic labels. For simplifying learning, we may only obtain the truth functions of some atomic label. For a given label, instances are divided into three kinds (positive, negative, and unclear) instead of two kinds as in popular studies so that the problem with binary relevance is avoided. For each instance, the classifier selects a compound label with most semantic information or richest connotation. As a predictive model, the semantic channel does not change with the prior probability distribution (source) of instances. It still works when the source is changed. The classifier changes with the source, and hence can overcome class-imbalance problem. It is shown that the old population's increasing will change the classifier for label "Old" and has been impelling the semantic evolution of "Old". The CM iteration algorithm for unseen instance classification is introduced.

Statistical models (SMs) include ordinary regression, Bayesian regression, semiparametric models, generalized additive models, longitudinal models, time-to-event models, penalized regression, and others. Penalized regression includes ridge regression, lasso, and elastic net. Contrary to what some machine learning (ML) researchers believe, SMs easily allow for complexity (nonlinearity and second-order interactions) and an unlimited number of candidate features (if penalized maximum likelihood estimation or Bayesian models are used). It is especially easy, using regression splines, to allow every continuous predictor to have a smooth nonlinear effect. ML is taken to mean an algorithmic approach that does not use traditional identified statistical parameters, and for which a preconceived structure is not imposed on the relationships between predictors and outcomes. ML usually does not attempt to isolate the effect of any single variable.

Journal of Parallel and Distributed Computing 104, 19-35.

Partially-observable Markov decision processes (POMDPs) with discounted-sum payoff are a standard framework to model a wide range of problems related to decision making under uncertainty. Traditionally, the goal has been to obtain policies that optimize the expectation of the discounted-sum payoff. A key drawback of the expectation measure is that even low probability events with extreme payoff can significantly affect the expectation, and thus the obtained policies are not necessarily risk-averse. An alternate approach is to optimize the probability that the payoff is above a certain threshold, which allows obtaining risk-averse policies, but ignores optimization of the expectation. We consider the expectation optimization with probabilistic guarantee (EOPG) problem, where the goal is to optimize the expectation ensuring that the payoff is above a given threshold with at least a specified probability. We present several results on the EOPG problem, including the first algorithm to solve it.

In the last years, deep learning algorithms have highly revolutionized several areas including speech, image and natural language processing. The specific field of Machine Translation (MT) has not remained invariant. Integration of deep learning in MT varies from re-modeling existing features into standard statistical systems to the development of a new architecture. Among the different neural networks, research works use feedforward neural networks, recurrent neural networks and the encoder-decoder schema. These architectures are able to tackle challenges as having low-resources or morphology variations. This manuscript focuses on describing how these neural networks have been integrated to enhance different aspects and models from statistical MT, including language modeling, word alignment, translation, reordering, and rescoring. Then, we report the new neural MT approach together with a description of the foundational related works and recent approaches on using subword, characters and training with multilingual languages, among others. Finally, we include an analysis of the corresponding challenges and future work in using deep learning in MT.

Dialogue management (DM) decides the next action of a dialogue system according to the current dialogue state, and thus plays a central role in task-oriented dialogue systems. Since dialogue management requires to have access to not only local utterances, but also the global semantics of the entire dialogue session, modeling the long-range history information is a critical issue. To this end, we propose a novel Memory-Augmented Dialogue management model (MAD) which employs a memory controller and two additional memory structures, i.e., a slot-value memory and an external memory. The slot-value memory tracks the dialogue state by memorizing and updating the values of semantic slots (for instance, cuisine, price, and location), and the external memory augments the representation of hidden states of traditional recurrent neural networks through storing more context information. To update the dialogue state efficiently, we also propose slot-level attention on user utterances to extract specific semantic information for each slot. Experiments show that our model can obtain state-of-the-art performance and outperforms existing baselines.