Cheng Zhang

Yashas Annadani, Nick Pawlowski, Joel Jennings et al.

Bayesian causal discovery aims to infer the posterior distribution over causal models from observed data, quantifying epistemic uncertainty and benefiting downstream tasks. However, computational challenges arise due to joint inference over combinatorial space of Directed Acyclic Graphs (DAGs) and nonlinear functions. Despite recent progress towards efficient posterior inference over DAGs, existing methods are either limited to variational inference on node permutation matrices for linear causal models, leading to compromised inference accuracy, or continuous relaxation of adjacency matrices constrained by a DAG regularizer, which cannot ensure resulting graphs are DAGs. In this work, we introduce a scalable Bayesian causal discovery framework based on a combination of stochastic gradient Markov Chain Monte Carlo (SG-MCMC) and Variational Inference (VI) that overcomes these limitations. Our approach directly samples DAGs from the posterior without requiring any DAG regularization, simultaneously draws function parameter samples and is applicable to both linear and nonlinear causal models. To enable our approach, we derive a novel equivalence to the permutation-based DAG learning, which opens up possibilities of using any relaxed gradient estimator defined over permutations. To our knowledge, this is the first framework applying gradient-based MCMC sampling for causal discovery. Empirical evaluation on synthetic and real-world datasets demonstrate our approach's effectiveness compared to state-of-the-art baselines.

Longlin Yu, Cheng Zhang

Semi-implicit variational inference (SIVI) greatly enriches the expressiveness of variational families by considering implicit variational distributions defined in a hierarchical manner. However, due to the intractable densities of variational distributions, current SIVI approaches often use surrogate evidence lower bounds (ELBOs) or employ expensive inner-loop MCMC runs for unbiased ELBOs for training. In this paper, we propose SIVI-SM, a new method for SIVI based on an alternative training objective via score matching. Leveraging the hierarchical structure of semi-implicit variational families, the score matching objective allows a minimax formulation where the intractable variational densities can be naturally handled with denoising score matching. We show that SIVI-SM closely matches the accuracy of MCMC and outperforms ELBO-based SIVI methods in a variety of Bayesian inference tasks.

Longlin Yu, Tianyu Xie, Yu Zhu et al. · pku

Semi-implicit variational inference (SIVI) has been introduced to expand the analytical variational families by defining expressive semi-implicit distributions in a hierarchical manner. However, the single-layer architecture commonly used in current SIVI methods can be insufficient when the target posterior has complicated structures. In this paper, we propose hierarchical semi-implicit variational inference, called HSIVI, which generalizes SIVI to allow more expressive multi-layer construction of semi-implicit distributions. By introducing auxiliary distributions that interpolate between a simple base distribution and the target distribution, the conditional layers can be trained by progressively matching these auxiliary distributions one layer after another. Moreover, given pre-trained score networks, HSIVI can be used to accelerate the sampling process of diffusion models with the score matching objective. We show that HSIVI significantly enhances the expressiveness of SIVI on several Bayesian inference problems with complicated target distributions. When used for diffusion model acceleration, we show that HSIVI can produce high quality samples comparable to or better than the existing fast diffusion model based samplers with a small number of function evaluations on various datasets.

Marcus Klasson, Hedvig Kjellström, Cheng Zhang

Replay methods are known to be successful at mitigating catastrophic forgetting in continual learning scenarios despite having limited access to historical data. However, storing historical data is cheap in many real-world settings, yet replaying all historical data is often prohibited due to processing time constraints. In such settings, we propose that continual learning systems should learn the time to learn and schedule which tasks to replay at different time steps. We first demonstrate the benefits of our proposal by using Monte Carlo tree search to find a proper replay schedule, and show that the found replay schedules can outperform fixed scheduling policies when combined with various replay methods in different continual learning settings. Additionally, we propose a framework for learning replay scheduling policies with reinforcement learning. We show that the learned policies can generalize better in new continual learning scenarios compared to equally replaying all seen tasks, without added computational cost. Our study reveals the importance of learning the time to learn in continual learning, which brings current research closer to real-world needs.

Matthew Ashman, Chao Ma, Agrin Hilmkil et al.

Latent confounding has been a long-standing obstacle for causal reasoning from observational data. One popular approach is to model the data using acyclic directed mixed graphs (ADMGs), which describe ancestral relations between variables using directed and bidirected edges. However, existing methods using ADMGs are based on either linear functional assumptions or a discrete search that is complicated to use and lacks computational tractability for large datasets. In this work, we further extend the existing body of work and develop a novel gradient-based approach to learning an ADMG with non-linear functional relations from observational data. We first show that the presence of latent confounding is identifiable under the assumptions of bow-free ADMGs with non-linear additive noise models. With this insight, we propose a novel neural causal model based on autoregressive flows for ADMG learning. This not only enables us to determine complex causal structural relationships behind the data in the presence of latent confounding, but also estimate their functional relationships (hence treatment effects) simultaneously. We further validate our approach via experiments on both synthetic and real-world datasets, and demonstrate the competitive performance against relevant baselines.

Ruibo Tu, Kun Zhang, Hedvig Kjellström et al.

To determine causal relationships between two variables, approaches based on Functional Causal Models (FCMs) have been proposed by properly restricting model classes; however, the performance is sensitive to the model assumptions, which makes it difficult to use. In this paper, we provide a novel dynamical-system view of FCMs and propose a new framework for identifying causal direction in the bivariate case. We first show the connection between FCMs and optimal transport, and then study optimal transport under the constraints of FCMs. Furthermore, by exploiting the dynamical interpretation of optimal transport under the FCM constraints, we determine the corresponding underlying dynamical process of the static cause-effect pair data. It provides a new dimension for describing static causal discovery tasks while enjoying more freedom for modeling the quantitative causal influences. In particular, we show that Additive Noise Models (ANMs) correspond to volume-preserving pressureless flows. Consequently, based on their velocity field divergence, we introduce a criterion for determining causal direction. With this criterion, we propose a novel optimal transport-based algorithm for ANMs which is robust to the choice of models and extend it to post-nonlinear models. Our method demonstrated state-of-the-art results on both synthetic and causal discovery benchmark datasets.

Chao Ma, Cheng Zhang

Real-world datasets often have missing values associated with complex generative processes, where the cause of the missingness may not be fully observed. This is known as missing not at random (MNAR) data. However, many imputation methods do not take into account the missingness mechanism, resulting in biased imputation values when MNAR data is present. Although there are a few methods that have considered the MNAR scenario, their model's identifiability under MNAR is generally not guaranteed. That is, model parameters can not be uniquely determined even with infinite data samples, hence the imputation results given by such models can still be biased. This issue is especially overlooked by many modern deep generative models. In this work, we fill in this gap by systematically analyzing the identifiability of generative models under MNAR. Furthermore, we propose a practical deep generative model which can provide identifiability guarantees under mild assumptions, for a wide range of MNAR mechanisms. Our method demonstrates a clear advantage for tasks on both synthetic data and multiple real-world scenarios with MNAR data.

Julia Grosse, Cheng Zhang, Philipp Hennig

Exciting contemporary machine learning problems have recently been phrased in the classic formalism of tree search -- most famously, the game of Go. Interestingly, the state-space underlying these sequential decision-making problems often posses a more general latent structure than can be captured by a tree. In this work, we develop a probabilistic framework to exploit a search space's latent structure and thereby share information across the search tree. The method is based on a combination of approximate inference in jointly Gaussian models for the explored part of the problem, and an abstraction for the unexplored part that imposes a reduction of complexity ad hoc. We empirically find our algorithm to compare favorably to existing non-probabilistic alternatives in Tic-Tac-Toe and a feature selection application.

Hippolyt Ritter, Martin Kukla, Cheng Zhang et al.

Bayesian neural networks and deep ensembles represent two modern paradigms of uncertainty quantification in deep learning. Yet these approaches struggle to scale mainly due to memory inefficiency issues, since they require parameter storage several times higher than their deterministic counterparts. To address this, we augment the weight matrix of each layer with a small number of inducing weights, thereby projecting the uncertainty quantification into such low dimensional spaces. We further extend Matheron's conditional Gaussian sampling rule to enable fast weight sampling, which enables our inference method to maintain reasonable run-time as compared with ensembles. Importantly, our approach achieves competitive performance to the state-of-the-art in prediction and uncertainty estimation tasks with fully connected neural networks and ResNets, while reducing the parameter size to $\leq 24.3\%$ of that of a $single$ neural network.

Haiyan Yin, Yingzhen Li, Sinno Jialin Pan et al.

Solving real-life sequential decision making problems under partial observability involves an exploration-exploitation problem. To be successful, an agent needs to efficiently gather valuable information about the state of the world for making rewarding decisions. However, in real-life, acquiring valuable information is often highly costly, e.g., in the medical domain, information acquisition might correspond to performing a medical test on a patient. This poses a significant challenge for the agent to perform optimally for the task while reducing the cost for information acquisition. In this paper, we propose a model-based reinforcement learning framework that learns an active feature acquisition policy to solve the exploration-exploitation problem during its execution. Key to the success is a novel sequential variational auto-encoder that learns high-quality representations from partially observed states, which are then used by the policy to maximize the task reward in a cost efficient manner. We demonstrate the efficacy of our proposed framework in a control domain as well as using a medical simulator. In both tasks, our proposed method outperforms conventional baselines and results in policies with greater cost efficiency.

Hengyi Cai, Hongshen Chen, Yonghao Song et al.

Current state-of-the-art neural dialogue models learn from human conversations following the data-driven paradigm. As such, a reliable training corpus is the crux of building a robust and well-behaved dialogue model. However, due to the open-ended nature of human conversations, the quality of user-generated training data varies greatly, and effective training samples are typically insufficient while noisy samples frequently appear. This impedes the learning of those data-driven neural dialogue models. Therefore, effective dialogue learning requires not only more reliable learning samples, but also fewer noisy samples. In this paper, we propose a data manipulation framework to proactively reshape the data distribution towards reliable samples by augmenting and highlighting effective learning samples as well as reducing the effect of inefficient samples simultaneously. In particular, the data manipulation model selectively augments the training samples and assigns an importance weight to each instance to reform the training data. Note that, the proposed data manipulation framework is fully data-driven and learnable. It not only manipulates training samples to optimize the dialogue generation model, but also learns to increase its manipulation skills through gradient descent with validation samples. Extensive experiments show that our framework can improve the dialogue generation performance with respect to various automatic evaluation metrics and human judgments.

Hengyi Cai, Hongshen Chen, Cheng Zhang et al.

Current state-of-the-art neural dialogue systems are mainly data-driven and are trained on human-generated responses. However, due to the subjectivity and open-ended nature of human conversations, the complexity of training dialogues varies greatly. The noise and uneven complexity of query-response pairs impede the learning efficiency and effects of the neural dialogue generation models. What is more, so far, there are no unified dialogue complexity measurements, and the dialogue complexity embodies multiple aspects of attributes---specificity, repetitiveness, relevance, etc. Inspired by human behaviors of learning to converse, where children learn from easy dialogues to complex ones and dynamically adjust their learning progress, in this paper, we first analyze five dialogue attributes to measure the dialogue complexity in multiple perspectives on three publicly available corpora. Then, we propose an adaptive multi-curricula learning framework to schedule a committee of the organized curricula. The framework is established upon the reinforcement learning paradigm, which automatically chooses different curricula at the evolving learning process according to the learning status of the neural dialogue generation model. Extensive experiments conducted on five state-of-the-art models demonstrate its learning efficiency and effectiveness with respect to 13 automatic evaluation metrics and human judgments.

Cheng Zhang, Qiuchi Li, Lingyu Hua et al.

It is known that Recurrent Neural Networks (RNNs) can remember, in their hidden layers, part of the semantic information expressed by a sequence (e.g., a sentence) that is being processed. Different types of recurrent units have been designed to enable RNNs to remember information over longer time spans. However, the memory abilities of different recurrent units are still theoretically and empirically unclear, thus limiting the development of more effective and explainable RNNs. To tackle the problem, in this paper, we identify and analyze the internal and external factors that affect the memory ability of RNNs, and propose a Semantic Euclidean Space to represent the semantics expressed by a sequence. Based on the Semantic Euclidean Space, a series of evaluation indicators are defined to measure the memory abilities of different recurrent units and analyze their limitations. These evaluation indicators also provide a useful guidance to select suitable sequence lengths for different RNNs during training.

Hengyi Cai, Hongshen Chen, Cheng Zhang et al.

Neural conversation systems generate responses based on the sequence-to-sequence (SEQ2SEQ) paradigm. Typically, the model is equipped with a single set of learned parameters to generate responses for given input contexts. When confronting diverse conversations, its adaptability is rather limited and the model is hence prone to generate generic responses. In this work, we propose an {\bf Ada}ptive {\bf N}eural {\bf D}ialogue generation model, \textsc{AdaND}, which manages various conversations with conversation-specific parameterization. For each conversation, the model generates parameters of the encoder-decoder by referring to the input context. In particular, we propose two adaptive parameterization mechanisms: a context-aware and a topic-aware parameterization mechanism. The context-aware parameterization directly generates the parameters by capturing local semantics of the given context. The topic-aware parameterization enables parameter sharing among conversations with similar topics by first inferring the latent topics of the given context and then generating the parameters with respect to the distributional topics. Extensive experiments conducted on a large-scale real-world conversational dataset show that our model achieves superior performance in terms of both quantitative metrics and human evaluations.

Maximilian Igl, Kamil Ciosek, Yingzhen Li et al.

The ability for policies to generalize to new environments is key to the broad application of RL agents. A promising approach to prevent an agent's policy from overfitting to a limited set of training environments is to apply regularization techniques originally developed for supervised learning. However, there are stark differences between supervised learning and RL. We discuss those differences and propose modifications to existing regularization techniques in order to better adapt them to RL. In particular, we focus on regularization techniques relying on the injection of noise into the learned function, a family that includes some of the most widely used approaches such as Dropout and Batch Normalization. To adapt them to RL, we propose Selective Noise Injection (SNI), which maintains the regularizing effect the injected noise has, while mitigating the adverse effects it has on the gradient quality. Furthermore, we demonstrate that the Information Bottleneck (IB) is a particularly well suited regularization technique for RL as it is effective in the low-data regime encountered early on in training RL agents. Combining the IB with SNI, we significantly outperform current state of the art results, including on the recently proposed generalization benchmark Coinrun.

Wenbo Gong, Sebastian Tschiatschek, Richard Turner et al.

In this paper we introduce the ice-start problem, i.e., the challenge of deploying machine learning models when only little or no training data is initially available, and acquiring each feature element of data is associated with costs. This setting is representative for the real-world machine learning applications. For instance, in the health-care domain, when training an AI system for predicting patient metrics from lab tests, obtaining every single measurement comes with a high cost. Active learning, where only the label is associated with a cost does not apply to such problem, because performing all possible lab tests to acquire a new training datum would be costly, as well as unnecessary due to redundancy. We propose Icebreaker, a principled framework to approach the ice-start problem. Icebreaker uses a full Bayesian Deep Latent Gaussian Model (BELGAM) with a novel inference method. Our proposed method combines recent advances in amortized inference and stochastic gradient MCMC to enable fast and accurate posterior inference. By utilizing BELGAM's ability to fully quantify model uncertainty, we also propose two information acquisition functions for imputation and active prediction problems. We demonstrate that BELGAM performs significantly better than the previous VAE (Variational autoencoder) based models, when the data set size is small, using both machine learning benchmarks and real-world recommender systems and health-care applications. Moreover, based on BELGAM, Icebreaker further improves the performance and demonstrate the ability to use minimum amount of the training data to obtain the highest test time performance.

Cheng Zhang, Cengiz Öztireli, Stephan Mandt et al.

The convergence speed of stochastic gradient descent (SGD) can be improved by actively selecting mini-batches. We explore sampling schemes where similar data points are less likely to be selected in the same mini-batch. In particular, we prove that such repulsive sampling schemes lowers the variance of the gradient estimator. This generalizes recent work on using Determinantal Point Processes (DPPs) for mini-batch diversification (Zhang et al., 2017) to the broader class of repulsive point processes. We first show that the phenomenon of variance reduction by diversified sampling generalizes in particular to non-stationary point processes. We then show that other point processes may be computationally much more efficient than DPPs. In particular, we propose and investigate Poisson Disk sampling---frequently encountered in the computer graphics community---for this task. We show empirically that our approach improves over standard SGD both in terms of convergence speed as well as final model performance.

Cheng Zhang, Judith Butepage, Hedvig Kjellstrom et al.

Many modern unsupervised or semi-supervised machine learning algorithms rely on Bayesian probabilistic models. These models are usually intractable and thus require approximate inference. Variational inference (VI) lets us approximate a high-dimensional Bayesian posterior with a simpler variational distribution by solving an optimization problem. This approach has been successfully used in various models and large-scale applications. In this review, we give an overview of recent trends in variational inference. We first introduce standard mean field variational inference, then review recent advances focusing on the following aspects: (a) scalable VI, which includes stochastic approximations, (b) generic VI, which extends the applicability of VI to a large class of otherwise intractable models, such as non-conjugate models, (c) accurate VI, which includes variational models beyond the mean field approximation or with atypical divergences, and (d) amortized VI, which implements the inference over local latent variables with inference networks. Finally, we provide a summary of promising future research directions.

Robert Bamler, Cheng Zhang, Manfred Opper et al.

Black box variational inference (BBVI) with reparameterization gradients triggered the exploration of divergence measures other than the Kullback-Leibler (KL) divergence, such as alpha divergences. In this paper, we view BBVI with generalized divergences as a form of estimating the marginal likelihood via biased importance sampling. The choice of divergence determines a bias-variance trade-off between the tightness of a bound on the marginal likelihood (low bias) and the variance of its gradient estimators. Drawing on variational perturbation theory of statistical physics, we use these insights to construct a family of new variational bounds. Enumerated by an odd integer order $K$, this family captures the standard KL bound for $K=1$, and converges to the exact marginal likelihood as $K\to\infty$. Compared to alpha-divergences, our reparameterization gradients have a lower variance. We show in experiments on Gaussian Processes and Variational Autoencoders that the new bounds are more mass covering, and that the resulting posterior covariances are closer to the true posterior and lead to higher likelihoods on held-out data.

Cheng Zhang, Hedvig Kjellstrom, Stephan Mandt

We study a mini-batch diversification scheme for stochastic gradient descent (SGD). While classical SGD relies on uniformly sampling data points to form a mini-batch, we propose a non-uniform sampling scheme based on the Determinantal Point Process (DPP). The DPP relies on a similarity measure between data points and gives low probabilities to mini-batches which contain redundant data, and higher probabilities to mini-batches with more diverse data. This simultaneously balances the data and leads to stochastic gradients with lower variance. We term this approach Diversified Mini-Batch SGD (DM-SGD). We show that regular SGD and a biased version of stratified sampling emerge as special cases. Furthermore, DM-SGD generalizes stratified sampling to cases where no discrete features exist to bin the data into groups. We show experimentally that our method results more interpretable and diverse features in unsupervised setups, and in better classification accuracies in supervised setups.

Cheng Zhang, Hedvig Kjellstrom, Bo C. Bertilson

In this paper, we explore the possibility to apply machine learning to make diagnostic predictions using discomfort drawings. A discomfort drawing is an intuitive way for patients to express discomfort and pain related symptoms. These drawings have proven to be an effective method to collect patient data and make diagnostic decisions in real-life practice. A dataset from real-world patient cases is collected for which medical experts provide diagnostic labels. Next, we extend a factorized multimodal topic model, Inter-Battery Topic Model (IBTM), to train a system that can make diagnostic predictions given an unseen discomfort drawing. Experimental results show reasonable predictions of diagnostic labels given an unseen discomfort drawing. The positive result indicates a significant potential of machine learning to be used for parts of the pain diagnostic process and to be a decision support system for physicians and other health care personnel.

Cheng Zhang, Babak Shahbaba, Hongkai Zhao

Traditionally, the field of computational Bayesian statistics has been divided into two main subfields: variational methods and Markov chain Monte Carlo (MCMC). In recent years, however, several methods have been proposed based on combining variational Bayesian inference and MCMC simulation in order to improve their overall accuracy and computational efficiency. This marriage of fast evaluation and flexible approximation provides a promising means of designing scalable Bayesian inference methods. In this paper, we explore the possibility of incorporating variational approximation into a state-of-the-art MCMC method, Hamiltonian Monte Carlo (HMC), to reduce the required gradient computation in the simulation of Hamiltonian flow, which is the bottleneck for many applications of HMC in big data problems. To this end, we use a {\it free-form} approximation induced by a fast and flexible surrogate function based on single-hidden layer feedforward neural networks. The surrogate provides sufficiently accurate approximation while allowing for fast exploration of parameter space, resulting in an efficient approximate inference algorithm. We demonstrate the advantages of our method on both synthetic and real data problems.

Cheng Zhang, Babak Shahbaba, Hongkai Zhao

For big data analysis, high computational cost for Bayesian methods often limits their applications in practice. In recent years, there have been many attempts to improve computational efficiency of Bayesian inference. Here we propose an efficient and scalable computational technique for a state-of-the-art Markov Chain Monte Carlo (MCMC) methods, namely, Hamiltonian Monte Carlo (HMC). The key idea is to explore and exploit the structure and regularity in parameter space for the underlying probabilistic model to construct an effective approximation of its geometric properties. To this end, we build a surrogate function to approximate the target distribution using properly chosen random bases and an efficient optimization process. The resulting method provides a flexible, scalable, and efficient sampling algorithm, which converges to the correct target distribution. We show that by choosing the basis functions and optimization process differently, our method can be related to other approaches for the construction of surrogate functions such as generalized additive models or Gaussian process models. Experiments based on simulated and real data show that our approach leads to substantially more efficient sampling algorithms compared to existing state-of-the art methods.

Cheng Zhang, Hedvig Kjellstrom

The quality of life of many people could be improved by autonomous humanoid robots in the home. To function in the human world, a humanoid household robot must be able to locate itself and perceive the environment like a human; scene perception, object detection and segmentation, and object spatial localization in 3D are fundamental capabilities for such humanoid robots. This paper presents a 3D multi-class object detection and segmentation method. The contributions are twofold. Firstly, we present a multi-class detection method, where a minimal joint codebook is learned in a principled manner. Secondly, we incorporate depth information using RGB-D imagery, which increases the robustness of the method and gives the 3D location of objects -- necessary since the robot reasons in 3D space. Experiments show that the multi-class extension improves the detection efficiency with respect to the number of classes and the depth extension improves the detection robustness and give sufficient natural 3D location of the objects.