Stein variational inference (SVI) is a sample-based approximate Bayesian inference technique that generates a sample set by jointly optimizing the samples' locations to minimize an information-theoretic measure of discrepancy with the target probability distribution. SVI thus provides a fast and significantly more sample-efficient approach to Bayesian inference than traditional (random-sampling-based) alternatives. However, the optimization techniques employed in existing SVI methods struggle to address problems in which the target distribution is high-dimensional, poorly-conditioned, or non-convex, which severely limits the range of their practical applicability. In this paper, we propose a novel trust-region optimization approach for SVI that successfully addresses each of these challenges. Our method builds upon prior work in SVI by leveraging conditional independences in the target distribution (to achieve high-dimensional scaling) and second-order information (to address poor conditioning), while additionally providing an effective adaptive step control procedure, which is essential for ensuring convergence on challenging non-convex optimization problems. Experimental results show our method achieves superior numerical performance, both in convergence rate and sample accuracy, and scales better in high-dimensional distributions, than previous SVI techniques.
User behavior modeling -- which aims to extract user interests from behavioral data -- has shown great power in Click-through rate (CTR) prediction, a key component in recommendation systems. Recently, attention-based algorithms have become a promising direction, as attention mechanisms emphasize the relevant interactions from rich behaviors. However, the methods struggle to capture the preferences of tail users with sparse interaction histories. To address the problem, we propose a novel variational inference approach, namely Group Prior Sampler Variational Inference (GPSVI), which introduces group preferences as priors to refine latent user interests for tail users. In GPSVI, the extent of adjustments depends on the estimated uncertainty of individual preference modeling. In addition, We further enhance the expressive power of variational inference by a volume-preserving flow. An appealing property of the GPSVI method is its ability to revert to traditional attention for head users with rich behavioral data while consistently enhancing performance for long-tail users with sparse behaviors. Rigorous analysis and extensive experiments demonstrate that GPSVI consistently improves the performance of tail users. Moreover, online A/B testing on a large-scale real-world recommender system further confirms the effectiveness of our proposed approach.
In recent years, Full-Waveform Inversion (FWI) has been extensively used to derive high-resolution subsurface velocity models from seismic data. However, due to the nonlinearity and ill-posed nature of the problem, FWI requires a good starting model to avoid producing non-physical solutions. Moreover, conventional optimization methods fail to quantify the uncertainty associated with the recovered solution, which is critical for decision-making processes. Bayesian inference offers an alternative approach as it directly or indirectly evaluates the posterior probability density function. For example, Markov Chain Monte Carlo (MCMC) methods generate multiple sample chains to characterize the solution's uncertainty. Despite their ability to theoretically handle any form of distribution, MCMC methods require many sampling steps; this limits their usage in high-dimensional problems with computationally intensive forward modeling, as is the FWI case. Variational Inference (VI), on the other hand, provides an approximate solution to the posterior distribution in the form of a parametric or non-parametric proposal distribution. Among the various algorithms used in VI, Stein Variational Gradient Descent (SVGD) is recognized for its ability to iteratively refine a set of samples to approximate the target distribution. However, mode and variance-collapse issues affect SVGD in high-dimensional inverse problems. This study aims to improve the performance of SVGD within the context of FWI by utilizing, for the first time, an annealed variant of SVGD and combining it with a multi-scale strategy. Additionally, we demonstrate that Principal Component Analysis (PCA) can be used to evaluate the performance of the optimization process. Clustering techniques are also employed to provide more rigorous and meaningful statistical analysis of the particles in the presence of multi-modal distributions.
While deep learning has expanded the possibilities for highly expressive variational families, the practical benefits of these tools for variational inference (VI) are often limited by the minimization of the traditional Kullback-Leibler objective, which can yield suboptimal solutions. A major challenge in this context is \emph{mode collapse}: the phenomenon where a model concentrates on a few modes of the target distribution during training, despite being statistically capable of expressing them all. In this work, we carry a theoretical investigation of mode collapse for the gradient flow on Gaussian mixture models. We identify the key low-dimensional statistics characterizing the flow, and derive a closed set of low-dimensional equations governing their evolution. Leveraging this compact description, we show that mode collapse is present even in statistically favorable scenarios, and identify two key mechanisms driving it: mean alignment and vanishing weight. Our theoretical findings are consistent with the implementation of VI using normalizing flows, a class of popular generative models, thereby offering practical insights.
Recommender systems predict personalized item rankings based on user preference distributions derived from historical behavior data. Recently, diffusion models (DMs) have gained attention in recommendation for their ability to model complex distributions, yet current DM-based recommenders often rely on traditional objectives like mean squared error (MSE) or recommendation objectives, which are not optimized for personalized ranking tasks or fail to fully leverage DM's generative potential. To address this, we propose PreferDiff, a tailored optimization objective for DM-based recommenders. PreferDiff transforms BPR into a log-likelihood ranking objective and integrates multiple negative samples to better capture user preferences. Specifically, we employ variational inference to handle the intractability through minimizing the variational upper bound and replaces MSE with cosine error to improve alignment with recommendation tasks. Finally, we balance learning generation and preference to enhance the training stability of DMs. PreferDiff offers three key benefits: it is the first personalized ranking loss designed specifically for DM-based recommenders and it improves ranking and faster convergence by addressing hard negatives. We also prove that it is theoretically connected to Direct Preference Optimization which indicates that it has the potential to align user preferences in DM-based recommenders via generative modeling. Extensive experiments across three benchmarks validate its superior recommendation performance and commendable general sequential recommendation capabilities. Our codes are available at https://github.com/lswhim/PreferDiff.
Several popular variational bounds involving importance weighting ideas have been proposed to generalize and improve on the Evidence Lower BOund (ELBO) in the context of maximum likelihood optimization, such as the Importance Weighted Auto-Encoder (IWAE) and the Variational R\'enyi (VR) bounds. The methodology to learn the parameters of interest using these bounds typically amounts to running gradient-based variational inference algorithms that incorporate the reparameterization trick. However, the way the choice of the variational bound impacts the outcome of variational inference algorithms can be unclear. Recently, the VR-IWAE bound was introduced as a variational bound that unifies the ELBO, IWAE and VR bounds methodologies. In this paper, we provide two analyses for the reparameterized and doubly-reparameterized gradient estimators of the VR-IWAE bound, which reveal the advantages and limitations of these gradient estimators while enabling us to compare of the ELBO, IWAE and VR bounds methodologies. Our work advances the understanding of importance weighted variational inference methods and we illustrate our theoretical findings empirically.
Estimating an individual's counterfactual outcomes under interventions is a challenging task for traditional causal inference and supervised learning approaches when the outcome is high-dimensional (e.g. gene expressions, facial images) and covariates are relatively limited. In this case, to predict one's outcomes under counterfactual treatments, it is crucial to leverage individual information contained in the observed outcome in addition to the covariates. Prior works using variational inference in counterfactual generative modeling have been focusing on neural adaptations and model variants within the conditional variational autoencoder formulation, which we argue is fundamentally ill-suited to the notion of counterfactual in causal inference. In this work, we present a novel variational Bayesian causal inference framework and its theoretical backings to properly handle counterfactual generative modeling tasks, through which we are able to conduct counterfactual supervision end-to-end during training without any counterfactual samples, and encourage disentangled exogenous noise abduction that aids the correct identification of causal effect in counterfactual generations. In experiments, we demonstrate the advantage of our framework compared to state-of-the-art models in counterfactual generative modeling on multiple benchmarks.
Given an intractable target density $p$, variational inference (VI) attempts to find the best approximation $q$ from a tractable family $Q$. This is typically done by minimizing the exclusive Kullback-Leibler divergence, $\text{KL}(q||p)$. In practice, $Q$ is not rich enough to contain $p$, and the approximation is misspecified even when it is a unique global minimizer of $\text{KL}(q||p)$. In this paper, we analyze the robustness of VI to these misspecifications when $p$ exhibits certain symmetries and $Q$ is a location-scale family that shares these symmetries. We prove strong guarantees for VI not only under mild regularity conditions but also in the face of severe misspecifications. Namely, we show that (i) VI recovers the mean of $p$ when $p$ exhibits an \textit{even} symmetry, and (ii) it recovers the correlation matrix of $p$ when in addition~$p$ exhibits an \textit{elliptical} symmetry. These guarantees hold for the mean even when $q$ is factorized and $p$ is not, and for the correlation matrix even when~$q$ and~$p$ behave differently in their tails. We analyze various regimes of Bayesian inference where these symmetries are useful idealizations, and we also investigate experimentally how VI behaves in their absence.
The logit transform is arguably the most widely-employed link function beyond linear settings. This transformation routinely appears in regression models for binary data and provides, either explicitly or implicitly, a core building-block within state-of-the-art methodologies for both classification and regression. Its widespread use, combined with the lack of analytical solutions for the optimization of general losses involving the logit transform, still motivates active research in computational statistics. Among the directions explored, a central one has focused on the design of tangent lower bounds for logistic log-likelihoods that can be tractably optimized, while providing a tight approximation of these log-likelihoods. Although progress along these lines has led to the development of effective minorize-maximize (MM) algorithms for point estimation and coordinate ascent variational inference schemes for approximate Bayesian inference under several logit models, the overarching focus in the literature has been on tangent quadratic minorizers. In fact, it is still unclear whether tangent lower bounds sharper than quadratic ones can be derived without undermining the tractability of the resulting minorizer. This article addresses such a challenging question through the design and study of a novel piece-wise quadratic lower bound that uniformly improves any tangent quadratic minorizer, including the sharpest ones, while admitting a direct interpretation in terms of the classical generalized lasso problem. As illustrated in a ridge logistic regression, this unique connection facilitates more effective implementations than those provided by available piece-wise bounds, while improving the convergence speed of quadratic ones.
Generative Flow Networks (GFlowNets) are amortized inference models designed to sample from unnormalized distributions over composable objects, with applications in generative modeling for tasks in fields such as causal discovery, NLP, and drug discovery. Traditionally, the training procedure for GFlowNets seeks to minimize the expected log-squared difference between a proposal (forward policy) and a target (backward policy) distribution, which enforces certain flow-matching conditions. While this training procedure is closely related to variational inference (VI), directly attempting standard Kullback-Leibler (KL) divergence minimization can lead to proven biased and potentially high-variance estimators. Therefore, we first review four divergence measures, namely, Renyi-$\alpha$'s, Tsallis-$\alpha$'s, reverse and forward KL's, and design statistically efficient estimators for their stochastic gradients in the context of training GFlowNets. Then, we verify that properly minimizing these divergences yields a provably correct and empirically effective training scheme, often leading to significantly faster convergence than previously proposed optimization. To achieve this, we design control variates based on the REINFORCE leave-one-out and score-matching estimators to reduce the variance of the learning objectives' gradients. Our work contributes by narrowing the gap between GFlowNets training and generalized variational approximations, paving the way for algorithmic ideas informed by the divergence minimization viewpoint.
Stein Variational Gradient Descent (SVGD) is a highly efficient method to sample from an unnormalized probability distribution. However, the SVGD update relies on gradients of the log-density, which may not always be available. Existing gradient-free versions of SVGD make use of simple Monte Carlo approximations or gradients from surrogate distributions, both with limitations. To improve gradient-free Stein variational inference, we combine SVGD steps with evolution strategy (ES) updates. Our results demonstrate that the resulting algorithm generates high-quality samples from unnormalized target densities without requiring gradient information. Compared to prior gradient-free SVGD methods, we find that the integration of the ES update in SVGD significantly improves the performance on multiple challenging benchmark problems.
Semantic communication technology emerges as a pivotal bridge connecting AI with classical communication. The current semantic communication systems are generally modeled as an Auto-Encoder (AE). AE lacks a deep integration of AI principles with communication strategies due to its inability to effectively capture channel dynamics. This gap makes it difficult to justify the need for joint source-channel coding (JSCC) and to explain why performance improves. This paper begins by exploring lossless and lossy communication, highlighting that the inclusion of data distortion distinguishes semantic communication from classical communication. It breaks the conditions for the separation theorem to hold and explains why the amount of data transferred by semantic communication is less. Therefore, employing JSCC becomes imperative for achieving optimal semantic communication. Moreover, a Variational Source-Channel Coding (VSCC) method is proposed for constructing semantic communication systems based on data distortion theory, integrating variational inference and channel characteristics. Using a deep learning network, we develop a semantic communication system employing the VSCC method and demonstrate its capability for semantic transmission. We also establish semantic communication systems of equivalent complexity employing the AE method and the VAE method. Experimental results reveal that the VSCC model offers superior interpretability compared to AE model, as it clearly captures the semantic features of the transmitted data, represented as the variance of latent variables in our experiments. In addition, VSCC model exhibits superior semantic transmission capabilities compared to VAE model. At the same level of data distortion evaluated by PSNR, VSCC model exhibits stronger human interpretability, which can be partially assessed by SSIM.
In sequential decision-making (SDM) tasks, methods like reinforcement learning (RL) and heuristic search have made notable advances in specific cases. However, they often require extensive exploration and face challenges in generalizing across diverse environments due to their limited grasp of the underlying decision dynamics. In contrast, large language models (LLMs) have recently emerged as powerful general-purpose tools, due to their capacity to maintain vast amounts of domain-specific knowledge. To harness this rich prior knowledge for efficiently solving complex SDM tasks, we propose treating LLMs as prior action distributions and integrating them into RL frameworks through Bayesian inference methods, making use of variational inference and direct posterior sampling. The proposed approaches facilitate the seamless incorporation of fixed LLM priors into both policy-based and value-based RL frameworks. Our experiments show that incorporating LLM-based action priors significantly reduces exploration and optimization complexity, substantially improving sample efficiency compared to traditional RL techniques, e.g., using LLM priors decreases the number of required samples by over 90% in offline learning scenarios.
Despite significant advancements in deep probabilistic models, learning low-dimensional discrete latent representations remains a challenging task. In this paper, we introduce a novel method that enhances variational inference in discrete latent variable models by leveraging Error Correcting Codes (ECCs) to introduce redundancy in the latent representations. This redundancy is then exploited by the variational posterior to yield more accurate estimates, thereby narrowing the variational gap. Inspired by ECCs commonly used in digital communications and data storage, we demonstrate proof-of-concept using a Discrete Variational Autoencoder (DVAE) with binary latent variables and block repetition codes. We further extend this idea to a hierarchical structure based on polar codes, where certain latent bits are more robustly protected. Our method improves generation quality, data reconstruction, and uncertainty calibration compared to the uncoded DVAE, even when trained with tighter bounds such as the Importance Weighted Autoencoder (IWAE) objective. In particular, we demonstrate superior performance on MNIST, FMNIST, CIFAR10, and Tiny ImageNet datasets. The general approach of integrating ECCs into variational inference is compatible with existing techniques to boost variational inference, such as importance sampling or Hamiltonian Monte Carlo. We also outline the key properties ECCs must have to effectively enhance discrete variational inference.
Value at Risk (VaR) and stress testing are two of the most widely used approaches in portfolio risk management to estimate potential market value losses under adverse market moves. VaR quantifies potential loss in value over a specified horizon (such as one day or ten days) at a desired confidence level (such as 95'th percentile). In scenario design and stress testing, the goal is to construct extreme market scenarios such as those involving severe recession or a specific event of concern (such as a rapid increase in rates or a geopolitical event), and quantify potential impact of such scenarios on the portfolio. The goal of this paper is to propose an approach for incorporating prevailing market conditions in stress scenario design and estimation of VaR so that they provide more accurate and realistic insights about portfolio risk over the near term. The proposed approach is based on historical data where historical observations of market changes are given more weight if a certain period in history is "more similar" to the prevailing market conditions. Clusters of market conditions are identified using a Machine Learning approach called Variational Inference (VI) where for each cluster future changes in portfolio value are similar. VI based algorithm uses optimization techniques to obtain analytical approximations of the posterior probability density of cluster assignments (market regimes) and probabilities of different outcomes for changes in portfolio value. Covid related volatile period around the year 2020 is used to illustrate the performance of the proposed approach and in particular show how VaR and stress scenarios adapt quickly to changing market conditions. Another advantage of the proposed approach is that classification of market conditions into clusters can provide useful insights about portfolio performance under different market conditions.
Many real-world datasets, such as healthcare, climate, and economics, are often collected as irregular time series, which poses challenges for accurate modeling. In this paper, we propose the Amortized Control of continuous State Space Model (ACSSM) for continuous dynamical modeling of time series for irregular and discrete observations. We first present a multi-marginal Doob's $h$-transform to construct a continuous dynamical system conditioned on these irregular observations. Following this, we introduce a variational inference algorithm with a tight evidence lower bound (ELBO), leveraging stochastic optimal control (SOC) theory to approximate the intractable Doob's $h$-transform and simulate the conditioned dynamics. To improve efficiency and scalability during both training and inference, ACSSM employs amortized inference to decouple representation learning from the latent dynamics. Additionally, it incorporates a simulation-free latent dynamics framework and a transformer-based data assimilation scheme, facilitating parallel inference of the latent states and ELBO computation. Through empirical evaluations across a variety of real-world datasets, ACSSM demonstrates superior performance in tasks such as classification, regression, interpolation, and extrapolation, while maintaining computational efficiency.
Existing diffusion-based methods for inverse problems sample from the posterior using score functions and accept the generated random samples as solutions. In applications that posterior mean is preferred, we have to generate multiple samples from the posterior which is time-consuming. In this work, by analyzing the probability density evolution of the conditional reverse diffusion process, we prove that the posterior mean can be achieved by tracking the mean of each reverse diffusion step. Based on that, we establish a framework termed reverse mean propagation (RMP) that targets the posterior mean directly. We show that RMP can be implemented by solving a variational inference problem, which can be further decomposed as minimizing a reverse KL divergence at each reverse step. We further develop an algorithm that optimizes the reverse KL divergence with natural gradient descent using score functions and propagates the mean at each reverse step. Experiments demonstrate the validity of the theory of our framework and show that our algorithm outperforms state-of-the-art algorithms on reconstruction performance with lower computational complexity in various inverse problems.
Obtaining a reliable estimate of the joint probability mass function (PMF) of a set of random variables from observed data is a significant objective in statistical signal processing and machine learning. Modelling the joint PMF as a tensor that admits a low-rank canonical polyadic decomposition (CPD) has enabled the development of efficient PMF estimation algorithms. However, these algorithms require the rank (model order) of the tensor to be specified beforehand. In real-world applications, the true rank is unknown. Therefore, an appropriate rank is usually selected from a candidate set either by observing validation errors or by computing various likelihood-based information criteria, a procedure which is computationally expensive for large datasets. This paper presents a novel Bayesian framework for estimating the joint PMF and automatically inferring its rank from observed data. We specify a Bayesian PMF estimation model and employ appropriate prior distributions for the model parameters, allowing for tuning-free rank inference via a single training run. We then derive a deterministic solution based on variational inference (VI) to approximate the posterior distributions of various model parameters. Additionally, we develop a scalable version of the VI-based approach by leveraging stochastic variational inference (SVI) to arrive at an efficient algorithm whose complexity scales sublinearly with the size of the dataset. Numerical experiments involving both synthetic data and real movie recommendation data illustrate the advantages of our VI and SVI-based methods in terms of estimation accuracy, automatic rank detection, and computational efficiency.
Variational inference in Bayesian deep learning often involves computing the gradient of an expectation that lacks a closed-form solution. In these cases, pathwise and score-function gradient estimators are the most common approaches. The pathwise estimator is often favoured for its substantially lower variance compared to the score-function estimator, which typically requires variance reduction techniques. However, recent research suggests that even pathwise gradient estimators could benefit from variance reduction. In this work, we review existing control-variates-based variance reduction methods for pathwise gradient estimators to assess their effectiveness. Notably, these methods often rely on integrand approximations and are applicable only to simple variational families. To address this limitation, we propose applying zero-variance control variates to pathwise gradient estimators. This approach offers the advantage of requiring minimal assumptions about the variational distribution, other than being able to sample from it.
Entity alignment (EA) aims to merge two knowledge graphs (KGs) by identifying equivalent entity pairs. Existing methods can be categorized into symbolic and neural models. Symbolic models, while precise, struggle with substructure heterogeneity and sparsity, whereas neural models, although effective, generally lack interpretability and cannot handle uncertainty. We propose NeuSymEA, a probabilistic neuro-symbolic framework that combines the strengths of both methods. NeuSymEA models the joint probability of all possible pairs' truth scores in a Markov random field, regulated by a set of rules, and optimizes it with the variational EM algorithm. In the E-step, a neural model parameterizes the truth score distributions and infers missing alignments. In the M-step, the rule weights are updated based on the observed and inferred alignments. To facilitate interpretability, we further design a path-ranking-based explainer upon this framework that generates supporting rules for the inferred alignments. Experiments on benchmarks demonstrate that NeuSymEA not only significantly outperforms baselines in terms of effectiveness and robustness, but also provides interpretable results.