Bayesian inference plays an important role in advancing machine learning, but faces computational challenges when applied to complex models such as deep neural networks. Variational inference circumvents these challenges by formulating Bayesian inference as an optimization problem and solving it using gradient-based optimization. In this paper, we argue in favor of natural-gradient approaches which, unlike their gradient-based counterparts, can improve convergence by exploiting the information geometry of the solutions. We show how to derive fast yet simple natural-gradient updates by using a duality associated with exponential-family distributions. An attractive feature of these methods is that, by using natural-gradients, they are able to extract accurate local approximations for individual model components. We summarize recent results for Bayesian deep learning showing the superiority of natural-gradient approaches over their gradient counterparts.

Wi-Fi) and the second/third/fourth generation (2/3/4G) mobile network, the number of mobile phones, which is 7.74 billion, 103.5 per 100 inhabitants all over the world in 2017, is rising dramatically [1]. Nowadays, mobile phone can not only send voice and text messages, but also easily and conveniently access the Internet which has been recognized as the most revolutionary development of Mobile Internet (M-Internet). Meanwhile, worldwide active mobile-broadband subscriptions in 2017 have increased to 4.22 billion, which is 9.21% higher than that in 2016 [1]. Figure 1 shows the numbers of mobile-cellular telephone and active mobile-broadband subscriptions of the world and main districts from 2010 to 2017. The numbers which are up to the bars are the mobile-cellular telephone or active mobile-broadband subscriptions (million) in the world of the year which increase each year. Under the M-Internet, various kinds of content (image, voice, video, etc.) can be sent and received everywhere and the related applications emerge to satisfy people's requirements, including working, study, daily life, entertainment, education, healthcare, etc. In China, mobile applications giants, i.e., Baidu, Alibaba and Tencent, held 78% of M-Internet online time per day in App which was about 2,412 minutes in 2017 [2]. This figure indicates that M-Internet has entered a rapidly growth stage.

We present an approximate Bayesian inference approach for estimating the intensity of a inhomogeneous Poisson process, where the intensity function is modelled using a Gaussian process (GP) prior via a sigmoid link function. Augmenting the model using a latent marked Poisson process and P\'olya--Gamma random variables we obtain a representation of the likelihood which is conjugate to the GP prior. We approximate the posterior using a free--form mean field approximation together with the framework of sparse GPs. Furthermore, as alternative approximation we suggest a sparse Laplace approximation of the posterior, for which an efficient expectation--maximisation algorithm is derived to find the posterior's mode. Results of both algorithms compare well with exact inference obtained by a Markov Chain Monte Carlo sampler and standard variational Gauss approach, while being one order of magnitude faster.

When learning from a batch of logged bandit feedback, the discrepancy between the policy to be learned and the off-policy training data imposes statistical and computational challenges. Unlike classical supervised learning and online learning settings, in batch contextual bandit learning, one only has access to a collection of logged feedback from the actions taken by a historical policy, and expect to learn a policy that takes good actions in possibly unseen contexts. Such a batch learning setting is ubiquitous in online and interactive systems, such as ad platforms and recommendation systems. Existing approaches based on inverse propensity weights, such as Inverse Propensity Scoring (IPS) and Policy Optimizer for Exponential Models (POEM), enjoy unbiasedness but often suffer from large mean squared error. In this work, we introduce a new approach named Maximum Likelihood Inverse Propensity Scoring (MLIPS) for batch learning from logged bandit feedback. Instead of using the given historical policy as the proposal in inverse propensity weights, we estimate a maximum likelihood surrogate policy based on the logged action-context pairs, and then use this surrogate policy as the proposal. We prove that MLIPS is asymptotically unbiased, and moreover, has a smaller nonasymptotic mean squared error than IPS. Such an error reduction phenomenon is somewhat surprising as the estimated surrogate policy is less accurate than the given historical policy. Results on multi-label classification problems and a large- scale ad placement dataset demonstrate the empirical effectiveness of MLIPS. Furthermore, the proposed surrogate policy technique is complementary to existing error reduction techniques, and when combined, is able to consistently boost the performance of several widely used approaches.

Personal use of this material is permitted. Abstract--This work presents an efficient framework to generate a motion plan of a robot with high degrees of freedom (e.g., a humanoid robot). High-dimensionality of the robot configuration space often leads to difficulties in utilizing the widely-used motion planning algorithms, since the volume of the decision space increases exponentially with the number of dimensions. T o handle complications arising from the large decision space, and to solve a corresponding motion planning problem efficiently, two key concepts are adopted in this work: First, the Gaussian process latent variable model (GP-L VM) is utilized for low-dimensional representation of the original configuration space. Second, an approximate inference algorithm is used, exploiting through the duality between control and estimation, to explore the decision space and to compute a high-quality motion trajectory of the robot. Utilizing the GP-L VM and the duality between control and estimation, we construct a fully probabilistic generative model with which a high-dimensional motion planning problem is transformed into a tractable inference problem. Finally, we compute the motion trajectory via an approximate inference algorithm based on a variant of the particle filter . OR robotic motion planning, a trajectory is designed for robot states through a complex configuration space from an initial state to perform a given task. The planning problem is formulated as an optimal control (OC) problem considering the robot dynamics for a feasible motion trajectory and the cost function for the task.

When presented with an unknown robot (subject) how can an autonomous agent (learner) figure out what this new robot can do? The subject's appearance can provide cues to its physical as well as cognitive capabilities. Seeing a humanoid can make one wonder if it can kick balls, climb stairs or recognize faces. What if the learner can request the subject to perform these tasks? We present an approach to make the learner build a model of the subject at a task based on the latter's appearance and refine it by experimentation. Apart from the subject's inherent capabilities, certain extrinsic factors may affect its performance at a task. Based on the subject's appearance and prior knowledge about the task a learner can identify a set of potential factors, a subset of which we assume are controllable. Our approach picks values of controllable factors to generate the most informative experiments to test the subject at. Additionally, we present a metric to determine if a factor should be incorporated in the model. We present results of our approach on modeling a humanoid robot at the task of kicking a ball. Firstly, we show that actively picking values for controllable factors, even in noisy experiments, leads to faster learning of the subject's model for the task. Secondly, starting from a minimal set of factors our metric identifies the set of relevant factors to incorporate in the model. Lastly, we show that the refined model better represents the subject's performance at the task.

Feature selection and attribute reduction are crucial problems, and widely used techniques in the field of machine learning, data mining and pattern recognition to overcome the well-known phenomenon of the Curse of Dimensionality, by either selecting a subset of features or removing unrelated ones. This paper presents a new feature selection method that efficiently carries out attribute reduction, thereby selecting the most informative features of a dataset. It consists of two components: 1) a measure for feature subset evaluation, and 2) a search strategy. For the evaluation measure, we have employed the fuzzy-rough dependency degree (FRFDD) in the lower approximation-based fuzzy-rough feature selection (L-FRFS) due to its effectiveness in feature selection. As for the search strategy, a new version of a binary shuffled frog leaping algorithm is proposed (B-SFLA). The new feature selection method is obtained by hybridizing the B-SFLA with the FRDD. Non-parametric statistical tests are conducted to compare the proposed approach with several existing methods over twenty two datasets, including nine high dimensional and large ones, from the UCI repository. The experimental results demonstrate that the B-SFLA approach significantly outperforms other metaheuristic methods in terms of the number of selected features and the classification accuracy.

We consider Markov models of stochastic processes where the next-step conditional distribution is defined by a kernel density estimator (KDE), similar to Markov forecast densities and certain time-series bootstrap schemes. The KDE Markov models (KDE-MMs) we discuss are nonlinear, nonparametric, fully probabilistic representations of stationary processes, based on techniques with strong asymptotic consistency properties. The models generate new data by concatenating points from the training data sequences in a context-sensitive manner, together with some additive driving noise. We present novel EM-type maximum-likelihood algorithms for data-driven bandwidth selection in KDE-MMs. Additionally, we augment the KDE-MMs with a hidden state, yielding a new model class, KDE-HMMs. The added state variable captures non-Markovian long memory and signal structure (e.g., slow oscillations), complementing the short-range dependences described by the Markov process. The resulting joint Markov and hidden-Markov structure is appealing for modelling complex real-world processes such as speech signals. We present guaranteed-ascent EM-update equations for model parameters in the case of Gaussian kernels, as well as relaxed update formulas that greatly accelerate training in practice. Experiments demonstrate increased held-out set probability for KDE-HMMs on several challenging natural and synthetic data series, compared to traditional techniques such as autoregressive models, HMMs, and their combinations.

Generating versatile and appropriate synthetic speech requires control over the output expression separate from the spoken text. Important non-textual speech variation is seldom annotated, in which case output control must be learned in an unsupervised fashion. In this paper, we perform an in-depth study of methods for unsupervised learning of control in statistical speech synthesis. For example, we show that popular unsupervised training heuristics can be interpreted as variational inference in certain autoencoder models. We additionally connect these models to VQ-VAEs, another, recently-proposed class of deep variational autoencoders, which we show can be derived from a very similar mathematical argument. The implications of these new probabilistic interpretations are discussed. We illustrate the utility of the various approaches with an application to emotional speech synthesis, where the unsupervised methods for learning expression control (without access to emotional labels) are found to give results that in many aspects match or surpass the previous best supervised approach.

We describe a general framework for probabilistic modeling of complex scenes and inference from ambiguous observations. The approach is motivated by applications in image analysis and is based on the use of priors defined by stochastic grammars. We define a class of grammars that capture relationships between the objects in a scene and provide important contextual cues for statistical inference. The distribution over scenes defined by a probabilistic scene grammar can be represented by a graphical model and this construction can be used for efficient inference with loopy belief propagation. We show experimental results with two different applications. One application involves the reconstruction of binary contour maps. Another application involves detecting and localizing faces in images. In both applications the same framework leads to robust inference algorithms that can effectively combine local information to reason about a scene.