UAI 2025 Papers

Optimizing expensive black-box functions under unknown constraints is a fundamental challenge across a range of real-world domains, such as hyperparameter tuning in machine learning, safe control in robotics, and material or drug discovery. In these settings, each function evaluation may be costly or time-consuming, and the system may need to operate within unknown or difficult-to-specify safety boundaries. We apply the Expected Improvement (EI) acquisition function to select the next samples within a feasible region, determined by Lower Confidence Bound (LCB) conditions for all constraints. The LCB approach guarantees constraint feasibility, while EI efficiently balances exploration and exploitation, especially when the feasible regions are much smaller than the overall search space. To model both the objective function and constraints, we use Deep Neural Networks (DNNs) instead of Gaussian Processes (GPs) to improve scalability and handle complex structured data. We provide a theoretical analysis showing our method’s convergence using recent Neural Tangent Kernel (NTK) theory. Under regularity conditions, both cumulative regret and constraint violation are bounded by the maximum information gain, with equivalent upper bounds to GP-based methods. To validate our algorithm, we conduct experiments on synthetic and real-world benchmarks, showing its benefit over recent methods in black-box optimization with unknown constraints.

Modeling hidden factors driving user preferences is crucial for recommendation yet challenging due to sparse rating data. While aligning preference factors from ratings and texts, as a solution, shows improvements, existing methods impose restrictive one-to-one factor correspondences and underutilize cross-modal interest signals. We propose an optimal transport (OT) approach to address these gaps. By modeling rating- and text-based preference factors as distributions, we compute an OT plan that captures their probabilistic relationships. This plan serves dual roles: 1) to regularize cross-modal preference factors without rigid correspondence assumptions, and 2) to blend preference signals across modalities through barycentric mapping. Experiments on real-world datasets validate our method’s effectiveness over competitive baselines, highlighting its novel use of OT for adaptive preference factor alignment, an underexplored direction in recommender system research.

We introduce HDP-Flow, a Bayesian nonparametric (BNP) model for unsupervised state discovery in dynamic, non-stationary time series data. Unlike prior work that assumes fixed states, HDPFlow models evolving datasets with unknown and variable latent states. By integrating the adaptability of BNP models with the expressive power of normalizing flows, HDP-Flow effectively models dynamic, non-stationary patterns, while learning transferable states across datasets with wellcalibrated uncertainty. We propose a scalable variational algorithm to enable efficient inference, addressing the limitations of traditional sampling-based BNP methods. HDP-Flow outperforms existing approaches in latent state identification and provides probabilistic insight into state distributions and transition dynamics. Evaluating HDP-Flow across two wearable datasets demonstrates transferability of states across diverse sub-populations, validating its robustness and generalizability.

Machine learning systems deployed in the real world must operate under dynamic and often unpredictable distribution shifts. This challenges the validity of statistical safety assurances on the system’s risk established beforehand. Common risk control frameworks rely on fixed assumptions and lack mechanisms to continuously monitor deployment reliability. In this work, we propose a general framework for the real-time monitoring of risk violations in evolving data streams. Leveraging the ‘testing by betting’ paradigm, we propose a sequential hypothesis testing procedure to detect violations of bounded risks associated with the model’s decision-making mechanism, while ensuring control on the false alarm rate. Our method operates under minimal assumptions on the nature of encountered shifts, rendering it broadly applicable. We illustrate the effectiveness of our approach by monitoring risks in outlier detection and set prediction under a variety of shifts.

The interactions between the update of model parameters and the update of perturbation operators complicate the dynamics of adversarial training (AT). This paper reveals a surprising behavior in AT, namely that the distribution induced by adversarial perturbations during AT becomes progressively more difficult to learn. We derived a generalization bound to theoretically attribute this behavior to the increasing of a quantity associated with the perturbation operator, namely, its local dispersion. We corroborate this explanation with concrete experimental validations and show that this deteriorating behavior of the induced distributions is correlated with robust overfitting of AT.

Learning effective data representations has been crucial in non-parametric two-sample testing. Common approaches will first split data into training and test sets and then learn data representations purely on the training set. However, recent theoretical studies have shown that, as long as the sample indexes are not used during the learning process, the whole data can be used to learn data representations, meanwhile ensuring control of Type-I errors. The above fact motivates us to use the test set (but without sample indexes) to facilitate the data representation learning in the testing. To this end, we propose a representation-learning two-sample testing (RL-TST) framework. RL-TST first performs purely self-supervised representation learning on the entire dataset to capture inherent representations (IRs) that reflect the underlying data manifold. A discriminative model is then trained on these IRs to learn discriminative representations (DRs), enabling the framework to leverage both the rich structural information from IRs and the discriminative power of DRs. Extensive experiments demonstrate that RL-TST outperforms representative approaches by simultaneously using data manifold information in the test set and enhancing test power via finding the DRs with the training set.

This paper investigates metric learning in a Reproducing Kernel Hilbert Space (RKHS) based on a set of random triplet comparisons in the form of *"Do you think item h is more similar to item i or item j?"* indicating similarity and differences between various items. The goal is to learn a metric in the RKHS that reflects the comparisons. Nonlinear metric learning using kernel methods and neural networks has shown great empirical promise. While previous works have addressed certain aspects of this problem, there is little or no theoretical understanding of such methods. The exception is the special (linear) case in which the RKHS is the standard d-dimensional Euclidean space; there is a comprehensive theory for metric learning in the d-dimensional Euclidean space. This paper develops a general RKHS framework for metric learning and provides novel generalization guarantees and sample complexity bounds. We validate our findings through a set of simulations and experiments on real datasets. Our code is publicly available at https://github.com/RamyaLab/metric-learning-RKHS.

Extracting meaningful latent representations from high-dimensional sequential data is a crucial challenge in machine learning, with applications spanning natural science and engineering. We introduce InfoDPCCA, a dynamic probabilistic Canonical Correlation Analysis (CCA) framework designed to model two interdependent sequences of observations. InfoDPCCA leverages a novel information-theoretic objective to extract a shared latent representation that captures the mutual structure between the data streams and balances representation compression and predictive sufficiency while also learning separate latent components that encode information specific to each sequence. Unlike prior dynamic CCA models, such as DPCCA, our approach explicitly enforces the shared latent space to encode only the mutual information between the sequences, improving interpretability and robustness. We further introduce a two-step training scheme to bridge the gap between information-theoretic representation learning and generative modeling, along with a residual connection mechanism to enhance training stability. Through experiments on synthetic and medical fMRI data, we demonstrate that InfoDPCCA excels as a tool for representation learning. Code of InfoDPCCA is available at https://github.com/marcusstang/InfoDPCCA.

With the extensive deployment and application of Internet of Things (IoT) devices, vulnerable edge nodes have emerged as primary targets for Advanced Persistent Threat (APT) attacks. Attackers compromise IoT terminal devices to establish an initial foothold and subsequently exploit lateral movement techniques to progressively infiltrate core business networks. Prior investigation methods struggle with fragmented threat intelligence and sparse attack samples in heterogeneous audit logs, resulting in incomplete attack chain reconstruction and high false positives. We propose a novel approach to APT attack investigation, FALCON, which captures complex causal relationships between entities from discrete audit logs and constructs cross-domain provenance graphs, enabling rapid and accurate identification of potential APT activities. FALCON trains an adaptive edge-side local model with cross-domain behavior sequences containing extensive and remote contextual information, and employs a bidirectional transformer pre-trained model to learn latent representations from unlabeled sequences. To the best of our knowledge, FALCON is the first APT investigation method to conduct causal provenance based on cross-domain audit logs while ensuring privacy protection. The experimental results demonstrate that FALCON effectively detects APT attacks with accuracy 99.71

When explaining machine learning models, it is important for explanations to have certain properties like faithfulness, robustness, smoothness, low complexity, etc. However, many properties are in tension with each other, making it challenging to achieve them simultaneously. For example, reducing the complexity of an explanation can make it less expressive, compromising its faithfulness. The ideal balance of trade-offs between properties tends to vary across different tasks and users. Motivated by these varying needs, we aim to find explanations that make optimal trade-offs while allowing for transparent control over the balance between different properties. Unlike existing methods that encourage desirable properties implicitly through their design, our approach optimizes explanations explicitly for a linear mixture of multiple properties. By adjusting the mixture weights, users can control the balance between those properties and create explanations with precisely what is needed for their particular task.

Human-robot collaboration (HRC) integrates the consistency and precision of robotic systems with the dexterity and cognitive abilities of humans to create synergy. However, human performance may degrade due to various factors (e.g., fatigue, trust) which can manifest unpredictably, and typically results in diminished output and reduced quality. To address this challenge toward successful HRCs, we present a human-aware approach to collaboration using a novel multi-agent decision-making framework. Type-based decentralized Markov decision processes (TB-DecMDP) additionally model latent, causal decision-making factors influencing agent behavior (e.g., fatigue), leading to dynamic agent types. In this framework, agents can switch between types and each maintains a belief about others’ current type based on observed actions while aiming to achieve a shared objective. We introduce a new inverse reinforcement learning (IRL) algorithm, TB-DecAIRL, which uses TB-DecMDP to model complex HRCs. TB-DecAIRL learns a type-contingent reward function and corresponding vector of policies from team demonstrations. Our evaluations in a realistic HRC problem setting establish that modeling human types in TB-DecAIRL improves robot behavior on the default of ignoring human factors, by increasing throughput in a human-robot produce sorting task.

Although there exist instance-dependent regret bounds for linear Markov decision processes (MDPs) and low-rank bandits, extensions to low-rank MDPs remain unexplored. In this work, we close this gap and provide regret bounds for low-rank MDPs in an instance-dependent setting. Specifically, we introduce an algorithm, called UniSREP-UCB, which utilizes a constrained optimization objective to learn features with good spectral properties. Furthermore, we demonstrate that our algorithm enjoys constant regret if the minimal sub-optimality gap and the occupancy distribution of the optimal policy are well-defined and known. To the best of our knowledge, these are the first instance-dependent regret results for low-rank MDPs.

In classification tasks, examples belonging to the same class can often still differ substantially from one another, and being able to capture such heterogeneity and its impact on classification can be important for aggregating estimates across multiple classifiers. Bayesian models developed so far have relied on a fixed set of latent variables to model these causal factors, which not only introduces the need for model selection but also assumes that each item is governed by the same set of causal factors. We develop a Bayesian model that can infer generic item features by modeling item feature membership as distributed according to an Indian Buffet Process. Despite its flexibility, our model is scalable to a large number of classifiers and examples. We compare our method with models from item response theory and Bayesian classifier combination on black-box crowdsourcing tasks and with neural network instance-dependent models in white-box classifier combination tasks.

Aggregate games model interdependent decision making when an agent’s utility depends on their own choice and the aggregation of everyone’s choices. We define a compactly representable subclass of aggregate games we call additive aggregate games, which encompasses popular games like congestion games, anonymous games, Schelling games, etc. We study computational questions on pure Nash equilibrium (PNE) and pure strong Nash equilibrium (SNE). We show that PNE existence is NP-complete for very simple cases of additive aggregate games. We devise an efficient algorithmic scheme for deciding the existence of a PNE and computing one (if it exists) for bounded aggregate space. We also give an approximation algorithm for a special type of additive aggregate games. For SNE, we show that SNE recognition is co-NP-complete and SNE existence is Σ^P_2-complete, even for simple types of additive aggregate games. For broad classes, we provide several novel and efficient aggregate-space algorithms for recognizing an SNE and deciding the existence of an SNE. Finally, we connect our results to several popular classes of games and show how our computational schemes can shed new light on these games.

Deep neural networks (DNNs) are highly susceptible to adversarial samples, raising concerns about their reliability in safety-critical tasks. Currently, methods of evaluating adversarial robustness are primarily categorized into attack-based and certified robustness evaluation approaches. The former not only relies on specific attack algorithms but also is highly time-consuming, while the latter due to its analytical nature, is typically difficult to implement for large and complex models. A few studies evaluate model robustness based on the model’s decision boundary, but they suffer from low evaluation accuracy. To address the aforementioned issues, we propose a novel adversarial robustness evaluation metric, Robustness Difference Index (RDI), which is based on model statistical features. RDI draws inspiration from clustering evaluation by analyzing the intra-class and inter-class distances of feature vectors separated by the decision boundary to quantify model robustness. It is attack-independent and has high computational efficiency. Experiments show that, RDI demonstrates a stronger correlation with the gold-standard adversarial robustness metric of attack success rate (ASR). The average computation time of RDI is only 1/30 of the evaluation method based on the PGD attack. Our open-source code is available at: https://github.com/BUPTAIOC/RDI.

Uncertainty-aware user modeling is crucial for designing AI systems that adapt to users in real-time while addressing privacy concerns. This paper proposes a novel framework for privacy-preserving probabilistic user modeling that integrates uncertainty quantification and differential privacy (DP). Building on neural processes (NPs), a scalable latent variable probabilistic model, we enable meta-learning for user behaviour prediction under privacy constraints. By employing differentially private stochastic gradient descent (DP-SGD), our method achieves rigorous privacy guarantees while preserving predictive accuracy. Unlike prior work, which primarily addresses privacy-preserving learning for convex or smooth functions, we establish theoretical guarantees for non-convex objectives, focusing on the utility-privacy trade-offs inherent in uncertainty-aware models. Through extensive experiments, we demonstrate that our approach achieves competitive accuracy under stringent privacy budgets. Our results showcase the potential of privacy-preserving probabilistic user models to enable trustworthy AI systems in real-world interactive applications.

Generalization outside the scope of one’s training data requires leveraging prior knowledge about the effects that transfer, and the effects that don’t, between different data sources. Transfer learning is a framework for specifying and refining this knowledge about sets of source (training) and target (prediction) data. A challenging open problem is addressing the empirical phenomenon of negative transfer, whereby the transfer learner performs worse on the target data after taking the source data into account than before. We first introduce a Bayesian perspective on negative transfer, and then a method to address it. The key insight from our formulation is that negative transfer can stem from misspecified prior information about non-transferable causes of the source data. Our proposed method, proxy-informed robust method for probabilistic transfer learning (PROMPT), does not require prior knowledge of the source data (the data sources may be "unknown"). PROMPT is thus applicable when differences between tasks are unobserved, such as in the presence of latent confounders. Moreover, the learner need not have access to observations in the target task (may not have the ability to "fine-tune"), and instead makes use of proxy (indirect) information. Our theoretical results show that the threat of negative transfer does not depend on the informativeness of the proxy information, highlighting the usefulness of PROMPT in cases where only noisy indirect information, such as human feedback, is available.

The machine learning community has recently put effort into quantized or low-precision arithmetics to scale large models. This paper proposes performing probabilistic inference in the quantized, discrete parameter space created by these representations, effectively enabling us to learn a continuous distribution using discrete parameters. We consider both 2D densities and quantized neural networks, where we introduce a tractable learning approach using probabilistic circuits. This method offers a scalable solution to manage complex distributions and provides clear insights into model behavior. We validate our approach with various models, demonstrating inference efficiency without sacrificing accuracy. This work advances scalable, interpretable machine learning by utilizing discrete approximations for probabilistic computations.

In many applications, especially due to lack of supervision or privacy concerns, the training data is grouped into bags of instances (feature-vectors) and for each bag we have only an aggregate label derived from the instance-labels in the bag. In learning from label proportions (LLP) the aggregate label is the average of the instance-labels in a bag, and a significant body of work has focused on training models in the LLP setting to predict instance-labels. In practice however, the training data may have fully supervised albeit covariate-shifted source data, along with the usual target data with bag-labels, and we wish to train a good instance-level predictor on the target domain. We call this the covariate-shifted hybrid LLP problem. Fully supervised covariate shifted data often has useful training signals and the goal is to leverage them for better predictive performance in the hybrid LLP setting. To achieve this, we develop methods for hybrid LLP which naturally incorporate the target bag-labels along with the source instance-labels, in the domain adaptation framework. Apart from proving theoretical guarantees bounding the target generalization error, we also conduct experiments on several publicly available datasets showing that our methods outperform LLP and domain adaptation baselines as well techniques from previous related work.