Bayesian inference methods for probabilistic topic models can quantify uncertainty in the parameters, which has primarily been used to increase the robustness of parameter estimates. In this work, we explore other rich information that can be obtained by analyzing the posterior distributions in topic models. Experimenting with latent Dirichlet allocation on two datasets, we propose ideas incorporating information about the posterior distributions at the topic level and at the word level. At the topic level, we propose a metric called topic stability that measures the variability of the topic parameters under the posterior. We show that this metric is correlated with human judgments of topic quality as well as with the consistency of topics appearing across multiple models. At the word level, we experiment with different methods for adjusting individual word probabilities within topics based on their uncertainty. Humans prefer words ranked by our adjusted estimates nearly twice as often when compared to the traditional approach. Finally, we describe how the ideas presented in this work could potentially applied to other predictive or exploratory models in future work.

Typically, neural conversation systems generate replies based on the sequence-to-sequence (seq2seq) model. seq2seq tends to produce safe and universal replies, which suffers from the lack of diversity and information. Determinantal Point Processes (DPPs) is a probabilistic model defined on item sets, which can select the items with good diversity and quality. In this paper, we investigate the diversity issue in two different aspects, namely query-level and system-level diversity. We propose a novel framework which organically combines seq2seq model with Determinantal Point Processes (DPPs). The new framework achieves high quality in generated reply and significantly improves the diversity among them. Experiments show that our model achieves the best performance among various baselines in terms of both quality and diversity.

Neural machine translation (NMT) heavily relies on parallel bilingual data for training. Since large-scale, high-quality parallel corpora are usually costly to collect, it is appealing to exploit monolingual corpora to improve NMT. Inspired by the law of total probability, which connects the probability of a given target-side monolingual sentence to the conditional probability of translating from a source sentence to the target one, we propose to explicitly exploit this connection to learn from and regularize the training of NMT models using monolingual data. The key technical challenge of this approach is that there are exponentially many source sentences for a target monolingual sentence while computing the sum of the conditional probability given each possible source sentence. We address this challenge by leveraging the dual translation model (target-to-source translation) to sample several mostly likely source-side sentences and avoid enumerating all possible candidate source sentences. That is, we transfer the knowledge contained in the dual model to boost the training of the primal model (source-to-target translation), and we call such an approach dual transfer learning. Experiment results on English-French and German-English tasks demonstrate that dual transfer learning achieves significant improvement over several strong baselines and obtains new state-of-the-art results.

Pre-trained distributed word representations have been proven to be useful in various natural language processing (NLP) tasks. However, the geometric basis of word representations and their relations to the representations of word's contexts has not been carefully studied yet. In this study, we first investigate such geometric relationship under a general framework, which is abstracted from some typical word representation learning approaches, and find out that only the directions of word representations are well associated to their context vector representations while the magnitudes are not. In order to make better use of the information contained in the magnitudes of word representations, we propose a hierarchical Gaussian model combined with maximum a posteriori estimation to learn word representations, and extend it to represent polysemous words. Our word representations have been evaluated on multiple NLP tasks, and the experimental results show that the proposed model achieved promising results, comparing to several popular word representations.

Dynamic topic models (DTM) are commonly used for mining latent topics in evolving web corpora. In this paper, we note that a major limitation of the conventional DTM based models is that they assume a predetermined and fixed scale of topics. In reality, however, topics may have varying spans and topics of multiple scales can co-exist in a single web or social media data stream. Therefore, DTMs that assume a fixed epoch length may not be able to effectively capture latent topics and thus negatively affect accuracy. In this paper, we propose a Multi-Scale Dynamic Topic Model (MS-DTM) and a complementary Incremental Multi-Scale Dynamic Topic Model (IMS-DTM) inference method that can be used to capture latent topics and their dynamics simultaneously, at different scales. In this model, topic specific feature distributions are generated based on a multi-scale feature distribution of the previous epochs; moreover, multiple scales of the current epoch are analyzed together through a novel multi-scale incremental Gibbs sampling technique. We show that the proposed model significantly improves efficiency and effectiveness compared to the single scale dynamic DTMs and prior models that consider only multiple scales of the past.

We introduce Graph-Structured Sum-Product Networks (GraphSPNs), a probabilistic approach to structured prediction for problems where dependencies between latent variables are expressed in terms of arbitrary, dynamic graphs. While many approaches to structured prediction place strict constraints on the interactions between inferred variables, many real-world problems can be only characterized using complex graph structures of varying size, often contaminated with noise when obtained from real data. Here, we focus on one such problem in the domain of robotics. We demonstrate how GraphSPNs can be used to bolster inference about semantic, conceptual place descriptions using noisy topological relations discovered by a robot exploring large-scale office spaces. Through experiments, we show that GraphSPNs consistently outperform the traditional approach based on undirected graphical models, successfully disambiguating information in global semantic maps built from uncertain, noisy local evidence. We further exploit the probabilistic nature of the model to infer marginal distributions over semantic descriptions of as yet unexplored places and detect spatial environment configurations that are novel and incongruent with the known evidence.

We present a probabilistic model for learning from dynamic relational data, wherein the observed interactions among networked nodes are modeled via the Bernoulli Poisson link function, and the underlying network structure are characterized by nonnegative latent node-group memberships, which are assumed to be gamma distributed. The latent memberships evolve according to Markov processes.The optimal number of latent groups can be determined by data itself. The computational complexity of our method scales with the number of non-zero links, which makes it scalable to large sparse dynamic relational data. We present batch and online Gibbs sampling algorithms to perform model inference. Finally, we demonstrate the model's performance on both synthetic and real-world datasets compared to state-of-the-art methods.

Recently, crowdsourcing has emerged as an effective paradigm for human-powered large scale problem solving in various domains. However, task requester usually has a limited amount of budget, thus it is desirable to have a policy to wisely allocate the budget to achieve better quality. In this paper, we study the principle of information maximization for active sampling strategies in the framework of HodgeRank, an approach based on Hodge Decomposition of pairwise ranking data with multiple workers. The principle exhibits two scenarios of active sampling: Fisher information maximization that leads to unsupervised sampling based on a sequential maximization of graph algebraic connectivity without considering labels; and Bayesian information maximization that selects samples with the largest information gain from prior to posterior, which gives a supervised sampling involving the labels collected. Experiments show that the proposed methods boost the sampling efficiency as compared to traditional sampling schemes and are thus valuable to practical crowdsourcing experiments.

Learning a function f(X) that predicts Y from X is the archetypal Machine Learning (ML) problem. Typically, both sets of attributes (i.e., X,Y) have to be known before a model can be trained. When this is not the case, or when functions f(X) that predict Y from X are needed for varying X and Y, this may introduce significant overhead (separate learning runs for each function). In this paper, we explore the possibility of omitting the specification of X and Y at training time altogether, by learning a multi-directional, or versatile model, which will allow prediction of any Y from any X. Specifically, we introduce a decision tree-based paradigm that generalizes the well-known Random Forests approach to allow for multi-directionality. The result of these efforts is a novel method called MERCS: Multi-directional Ensembles of Regression and Classification treeS. Experiments show the viability of the approach.

Bayesian optimization (BayesOpt) is a derivative-free approach for sequentially optimizing stochastic black-box functions. Standard BayesOpt, which has shown many successes in machine learning applications, assumes a finite dimensional domain which often is a parametric space. The parameter space is defined by the features used in the function approximations which are often selected manually. Therefore, the performance of BayesOpt inevitably depends on the quality of chosen features. This paper proposes a new Bayesian optimization framework that is able to optimize directly on the domain of function spaces. The resulting framework, Bayesian Functional Optimization (BFO), not only extends the application domains of BayesOpt to functional optimization problems but also relaxes the performance dependency on the chosen parameter space. We model the domain of functions as a reproducing kernel Hilbert space (RKHS), and use the notion of Gaussian processes on a real separable Hilbert space. As a result, we are able to define traditional improvement-based (PI and EI) and optimistic acquisition functions (UCB) as functionals. We propose to optimize the acquisition functionals using analytic functional gradients that are also proved to be functions in a RKHS. We evaluate BFO in three typical functional optimization tasks: i) a synthetic functional optimization problem, ii) optimizing activation functions for a multi-layer perceptron neural network, and iii) a reinforcement learning task whose policies are modeled in RKHS.