This paper studies a training method to jointly estimate an energy-based model and a flow-based model, in which the two models are iteratively updated based on a shared adversarial value function. This joint training method has the following traits. (1) The update of the energy-based model is based on noise contrastive estimation, with the flow model serving as a strong noise distribution. (2) The update of the flow model approximately minimizes the Jensen-Shannon divergence between the flow model and the data distribution. (3) Unlike generative adversarial networks (GAN) which estimates an implicit probability distribution defined by a generator model, our method estimates two explicit probabilistic distributions on the data. Using the proposed method we demonstrate a significant improvement on the synthesis quality of the flow model, and show the effectiveness of unsupervised feature learning by the learned energy-based model. Furthermore, the proposed training method can be easily adapted to semi-supervised learning. We achieve competitive results to the state-of-the-art semi-supervised learning methods.

Bayesian interpretations of neural network have a long history, dating back to early work in the 1990's and have recently regained attention because of their desirable properties like uncertainty estimation, model robustness and regularisation. We want to discuss here the application of Bayesian models to knowledge sharing between neural networks. Knowledge sharing comes in different facets, such as transfer learning, model distillation and shared embeddings. All of these tasks have in common that learned "features" ought to be shared across different networks. Theoretically rooted in the concepts of Bayesian neural networks this work has widespread application to general deep learning.

Child mortality from preventable diseases such as pneumonia and diarrhoea in low and middle-income countries remains a serious global challenge. We combine knowledge with available Demographic and Health Survey (DHS) data from India, to construct Bayesian Networks (BNs) and investigate the factors associated with childhood diarrhoea. We make use of freeware tools to learn the graphical structure of the DHS data with score-based, constraint-based, and hybrid structure learning algorithms. We investigate the effect of missing values, sample size, and knowledge-based constraints on each of the structure learning algorithms and assess their accuracy with multiple scoring functions. Weaknesses in the survey methodology and data available, as well as the variability in the BNs generated, mean that is not possible to learn a definitive causal BN from data. However, knowledge-based constraints are found to be useful in reducing the variation in the graphs produced by the different algorithms, and produce graphs which are more reflective of the likely influential relationships in the data. Furthermore, valuable insights are gained into the performance and characteristics of the structure learning algorithms. Two score-based algorithms in particular, TABU and FGES, demonstrate many desirable qualities; a) with sufficient data, they produce a graph which is similar to the reference graph, b) they are relatively insensitive to missing values, and c) behave well with knowledge-based constraints. The results provide a basis for further investigation of the DHS data and for a deeper understanding of the behaviour of the structure learning algorithms when applied to real-world settings.

Multimodal integration is an important process in perceptual decision-making. In humans, this process has often been shown to be statistically optimal, or near optimal: sensory information is combined in a fashion that minimises the average error in perceptual representation of stimuli. However, sometimes there are costs that come with the optimization, manifesting as illusory percepts. We review audio-visual facilitations and illusions that are products of multisensory integration, and the computational models that account for these phenomena. In particular, the same optimal computational model can lead to illusory percepts, and we suggest that more studies should be needed to detect and mitigate these illusions, as artefacts in artificial cognitive systems. We provide cautionary considerations when designing artificial cognitive systems with the view of avoiding such artefacts. Finally, we suggest avenues of research towards solutions to potential pitfalls in system design. We conclude that detailed understanding of multisensory integration and the mechanisms behind audio-visual illusions can benefit the design of artificial cognitive systems.

Adversarial Optimization (AO) provides a reliable, practical way to match two implicitly defined distributions, one of which is usually represented by a sample of real data, and the other is defined by a generator. Typically, AO involves training of a high-capacity model on each step of the optimization. In this work, we consider computationally heavy generators, for which training of high-capacity models is associated with substantial computational costs. To address this problem, we introduce a novel family of divergences, which varies the capacity of the underlying model, and allows for a significant acceleration with respect to the number of samples drawn from the generator. We demonstrate the performance of the proposed divergences on several tasks, including tuning parameters of a physics simulator, namely, Pythia event generator.

We study the problem of separating a mixture of distributions, all of which come from interventions on a known causal bayesian network. Given oracle access to marginals of all distributions resulting from interventions on the network, and estimates of marginals from the mixture distribution, we want to recover the mixing proportions of different mixture components. We show that in the worst case, mixing proportions cannot be identified using marginals only. If exact marginals of the mixture distribution were known, under a simple assumption of excluding a few distributions from the mixture, we show that the mixing proportions become identifiable. Our identifiability proof is constructive and gives an efficient algorithm recovering the mixing proportions exactly. When exact marginals are not available, we design an optimization framework to estimate the mixing proportions. Our problem is motivated from a real-world scenario of an e-commerce business, where multiple interventions occur at a given time, leading to deviations in expected metrics. We conduct experiments on the well known publicly available ALARM network and on a proprietary dataset from a large e-commerce company validating the performance of our method.

In my prior blog post, I wrote of a clever elf that could predict the outcome of a mathematically fair process roughly ninety percent of the time. Actually, it is ninety-three percent of the time and why it is ninety-three percent instead of ninety percent is also important. The purpose of the prior blog post was to illustrate the weakness of using the minimum variance unbiased estimator (MVUE) in applied finance. Nonetheless, that begs a more general question of when and why it should be used, or a Bayesian or Likelihood-based method should be applied. Fortunately, the prior blog post provides a way of looking at the problem. Fisher's Likelihood-based, Pearson and Neyman's Frequency-based and Laplace's method of inverse probability really are at odds with one another. Indeed, much of the literature of the mid-twentieth century had a polemical ring to it.

This paper proposes a formal approach to learning and planning for agents operating in a priori unknown, time-varying environments. The proposed method computes the maximally likely model of the environment, given the observations about the environment made by an agent earlier in the system run and assuming knowledge of a bound on the maximal rate of change of system dynamics. Such an approach generalizes the estimation method commonly used in learning algorithms for unknown Markov decision processes with time-invariant transition probabilities, but is also able to quickly and correctly identify the system dynamics following a change. Based on the proposed method, we generalize the exploration bonuses used in learning for time-invariant Markov decision processes by introducing a notion of uncertainty in a learned time-varying model, and develop a control policy for time-varying Markov decision processes based on the exploitation and exploration trade-off. We demonstrate the proposed methods on four numerical examples: a patrolling task with a change in system dynamics, a two-state MDP with periodically changing outcomes of actions, a wind flow estimation task, and a multi-arm bandit problem with periodically changing probabilities of different rewards.

Variational inference offers scalable and flexible tools to tackle intractable Bayesian inference of modern statistical models like Bayesian neural networks and Gaussian processes. For largely over-parameterized models, however, the over-regularization property of the variational objective makes the application of variational inference challenging. Inspired by the literature on kernel methods, and in particular on structured approximations of distributions of random matrices, this paper proposes Walsh-Hadamard Variational Inference, which uses Walsh-Hadamard-based factorization strategies to reduce model parameterization, accelerate computations, and increase the expressiveness of the approximate posterior beyond fully factorized ones.

It is well known that deep generative models have a rich latent space, and that it is possible to smoothly manipulate their outputs by traversing this latent space. Recently, architectures have emerged that allow for more complex manipulations, such as making an image look as though it were from a different class, or painted in a certain style. These methods typically require large amounts of training in order to learn a single class of manipulations. We present Transflow Learning, a method for transforming a pre-trained generative model so that its outputs more closely resemble data that we provide afterwards. In contrast to previous methods, Transflow Learning does not require any training at all, and instead warps the probability distribution from which we sample latent vectors using Bayesian inference. Transflow Learning can be used to solve a wide variety of tasks, such as neural style transfer and few-shot classification.