论文检索

Daniele Meli, Alberto Castellini, Alessandro Farinelli

Partially Observable Markov Decision Processes (POMDPs) are a powerful framework for planning under uncertainty. They allow to model state uncertainty as a belief probability distribution. Approximate solvers based on Monte Carlo sampling show great success to relax the computational demand and perform online planning. However, scaling to complex realistic domains with many actions and long planning horizons is still a major challenge, and a key point to achieve good performance is guiding the action-selection process with domain-dependent policy heuristics which are tailored for the specific application domain. We propose to learn high-quality heuristics from POMDP traces of executions generated by any solver. We convert the belief-action pairs to a logical semantics, and exploit data- and time-efficient Inductive Logic Programming (ILP) to generate interpretable belief-based policy specifications, which are then used as online heuristics. We evaluate thoroughly our methodology on two notoriously challenging POMDP problems, involving large action spaces and long planning horizons, namely, rocksample and pocman. Considering different state-of-the-art online POMDP solvers, including POMCP, DESPOT and AdaOPS, we show that learned heuristics expressed in Answer Set Programming (ASP) yield performance superior to neural networks and similar to optimal handcrafted task-specific heuristics within lower computational time. Moreover, they well generalize to more challenging scenarios not experienced in the training phase (e.g., increasing rocks and grid size in rocksample, incrementing the size of the map and the aggressivity of ghosts in pocman).

Daniel S. Brown

Ensuring that AI systems do what we, as humans, actually want them to do, is one of the biggest open research challenges in AI alignment and safety. My research seeks to directly address this challenge by enabling AI systems to interact with humans to learn aligned and robust behaviors. The way in which robots and other AI systems behave is often the result of optimizing a reward function. However, manually designing good reward functions is highly challenging and error prone, even for domain experts. Consider trying to write down a reward function that describes good driving behavior or how you like your bed made in the morning. While reward functions for these tasks are difficult to manually specify, human feedback in the form of demonstrations or preferences are often much easier to obtain. However, human data is often difficult to interpret, due to ambiguity and noise. Thus, it is critical that AI systems take into account epistemic uncertainty over the human's true intent. My talk will give an overview of my lab's progress along the following fundamental research areas: (1) efficiently maintaining uncertainty over human intent, (2) directly optimizing behavior to be robust to uncertainty over human intent, and (3) actively querying for additional human input to reduce uncertainty over human intent.

Niloufar Eghbali, Tuka Alhanai, Mohammad M. Ghassemi

Mechanical Ventilation (MV) is a critical life-support intervention in intensive care units (ICUs). However, optimal ventilator settings are challenging to determine because of the complexity of balancing patient-specific physiological needs with the risks of adverse outcomes that impact morbidity, mortality, and healthcare costs. This study introduces ConformalDQN, a novel distribution-free conformal deep Q-learning approach for optimizing mechanical ventilation in intensive care units. By integrating conformal prediction with deep reinforcement learning, our method provides reliable uncertainty quantification, addressing the challenges of Q-value overestimation and out-of-distribution actions in offline settings. We trained and evaluated our model using ICU patient records from the MIMIC-IV database. ConformalDQN extends the Double DQN architecture with a conformal predictor and employs a composite loss function that balances Q-learning with well-calibrated probability estimation. This enables uncertainty-aware action selection, allowing the model to avoid potentially harmful actions in unfamiliar states and handle distribution shifts by being more conservative in out-of-distribution scenarios. Evaluation against baseline models, including physician policies, policy constraint methods, and behavior cloning, demonstrates that ConformalDQN consistently makes recommendations within clinically safe and relevant ranges, outperforming other methods by increasing the 90-day survival rate. Notably, our approach provides an interpretable measure of confidence in its decisions, which is crucial for clinical adoption and potential human-in-the-loop implementations.

Tian Qin, Tian-Zuo Wang, Zhi-Hua Zhou

Machine learning (ML) has made significant advancements across various domains, with a shifting focus from purely predictive tasks to decision-making. The recent proposal by Zhou (2022) introduced a line of research known as rehearsal learning, which provides a novel perspective on modeling decision-making tasks. However, previous studies mainly focused on the linear Gaussian setting to constrain the modeling complexity. Furthermore, it has been demonstrated that finding exact optimal multivariate decisions within the sampling-based rehearsal framework is computationally infeasible in polynomial time, necessitating the development of approximate methods. In this work, we present Grad-Rh, the first gradient-based rehearsal learning method that can efficiently find multivariate decisions under non-linear and non-Gaussian settings. We address the uncertainty in decision-making tasks using flexible and expressive conditional normalizing flow models and derive four surrogate loss functions to enable efficient gradient-based optimization. Experimental results show that Grad-Rh performs comparably to exact baselines on linear data and significantly outperforms them on non-linear data in both decision quality and running time.

Haeun Jeon, Hyunglip Bae, Minsu Park, Chanyeong Kim, Woo Chang Kim

In decision-making problems under uncertainty, predicting unknown parameters is often considered independent of the optimization part. Decision-focused learning (DFL) is a task-oriented framework that integrates prediction and optimization by adapting the predictive model to give better decisions for the corresponding task. Here, an inevitable challenge arises when computing the gradients of the optimal decision with respect to the parameters. Existing research copes with this issue by smoothly reforming surrogate optimization or constructing surrogate loss functions that mimic task loss. However, they are applied to restricted optimization domains. In this paper, we propose Locally Convex Global Loss Network (LCGLN), a global surrogate loss model that can be implemented in a general DFL paradigm. LCGLN learns task loss via a partial input convex neural network which is guaranteed to be convex for chosen inputs while keeping the non-convex global structure for the other inputs. This enables LCGLN to admit general DFL through only a single surrogate loss without any sense for choosing appropriate parametric forms. We confirm the effectiveness and flexibility of LCGLN by evaluating our proposed model with three stochastic decision-making problems.

Martin Kurečka, Václav Nevyhoštěný, Petr Novotný, Vít Unčovský

Constrained Markov decision processes (CMDPs), in which the agent optimizes expected payoffs while keeping the expected cost below a given threshold, are the leading framework for safe sequential decision making under stochastic uncertainty. Among algorithms for planning and learning in CMDPs, methods based on Monte Carlo tree search (MCTS) have particular importance due to their efficiency and extendibility to more complex frameworks (such as partially observable settings and games). However, current MCTS-based methods for CMDPs either struggle with finding safe (i.e., constraint-satisfying) policies, or are too conservative and do not find valuable policies. We introduce Threshold UCT (T-UCT), an online MCTS-based algorithm for CMDP planning. Unlike previous MCTS-based CMDP planners, T-UCT explicitly estimates Pareto curves of cost-utility trade-offs throughout the search tree, using these together with a novel action selection and threshold update rule to seek safe and valuable policies. Our experiments demonstrate that our approach significantly outperforms state-of-the-art methods from the literature.

Li Huang, Haowen Liu, Qiang Gao, Jiajing Yu, Guisong Liu, Xueqin Chen

Few-shot Named Entity Recognition (NER) spotlights the tag of novel entity types in data-limited scenarios or lower-resource settings. Advances with Pre-trained Language Models (PLMs), including BERT, GPT, and their variants, have driven tremendous strategies to leverage context-dependent representations and exploit predefined relational cues, yielding significant gains in witnessing unseen entities. Nevertheless, a fundamental issue exists in prior efforts regarding their susceptibility to adversarial attacks in the intricate semantic environment. This vulnerability undermines the robustness of semantic representations, exacerbating the challenge of accurate entity identification, especially when transitioning across domains. To this end, we propose an Adversity-aware Augment Learning (AAL) solution for the few-shot NER task, dedicated to retrieving and reinforcing entity prototypes resilient to adversarial inference, thereby enhancing cross-domain semantic coherence. In particular, AAL employs a two-stage paradigm consisting of training and fine-tuning. The process initiates with augmentation learning by leveraging two kinds of prompt learning schemes, then identifies prototypes under the guidance of a variational manner. Furthermore, we devise a domain-oriented prototype refinement to optimize prototype learning under conditions of uncertainty attack, facilitating the effective transfer of common knowledge from source to target domains. The experimental results, encompassing the few-shot NER datasets under both certainty and uncertainty conditions, affirm the superiority of the proposed AAL over several representative baselines, particularly its capability against adversarial attacks.

Siyuan Zhang, Linbo Xie

Output uncertainty indicates whether the probabilistic properties of the overall distribution reflect objective characteristics of the model output. Unlike most loss functions and metrics in machine learning, uncertainty pertains to individual samples, but validating it on individual samples is unfeasible. When validated collectively, it cannot fully represent individual sample properties, posing a challenge in assessing and calibrating model confidence in a limited data set. Hence, it is crucial to consider confidence calibration characteristics. To counter the adverse effects of the gradual amplification of the classifier output amplitude in supervised learning, we introduce a post-processing parametric calibration method, ρ-Norm Scaling, which expands the calibrator expression and mitigates overconfidence due to excessive amplitude while preserving accuracy. Moreover, calibrator optimization based bin-level calibration error often results in the loss of significant instance-level information. Therefore, we include probability distribution regularization, which incorporates a priori information that the instance-level uncertainty distribution after calibration should resemble the distribution before calibration. Experimental results demonstrate the substantial enhancement in the post-processing calibrator for uncertainty calibration with our proposed method.

Shuo Sun, Meng Qi, Zuo-Jun Max Shen

In this work, we consider an online robust Markov Decision Process (MDP) where we have the information of finitely many prototypes of the underlying transition kernel. We consider an adaptively updated ambiguity set of the prototypes and propose an algorithm that efficiently identifies the true underlying transition kernel while guaranteeing the performance of the corresponding robust policy. To be more specific, we provide a sublinear regret of the subsequent optimal robust policy. We also provide an early stopping mechanism and a worst-case performance bound of the value function. In numerical experiments, we demonstrate that our method outperforms existing approaches, particularly in the early stage with limited data. This work contributes to robust MDPs by considering possible prior information about the underlying transition probability and online learning, offering both theoretical insights and practical algorithms for improved decision-making under uncertainty.

Hao Ma, Shijie Wang, Zhiqiang Pu, Siyao Zhao, Xiaolin Ai

Guiding the policy of multi-agent reinforcement learning to align with human common sense is a difficult problem, largely due to the complexity of modeling common sense as a reward, especially in complex and long-horizon multi-agent tasks. Recent works have shown the effectiveness of reward shaping, such as potential-based rewards, to enhance policy alignment. The existing works, however, primarily rely on experts to design rule-based rewards, which are often labor-intensive and lack a high-level semantic understanding of common sense. To solve this problem, we propose a hierarchical vision-based reward shaping method. At the bottom layer, a visual-language model (VLM) serves as a generic potential function, guiding the policy to align with human common sense through its intrinsic semantic understanding. To help the policy adapts to uncertainty and changes in long-horizon tasks, the top layer features an adaptive skill selection module based on a visual large language model (vLLM). The module uses instructions, video replays, and training records to dynamically select suitable potential function from a pre-designed pool. Besides, our method is theoretically proven to preserve the optimal policy. Extensive experiments conducted in the Google Research Football environment demonstrate that our method not only achieves a higher win rate but also effectively aligns the policy with human common sense.

Sanath Kumar Krishnamurthy, Tanmay Gangwani, Sumeet Katariya, Branislav Kveton, Shrey Modi, Anshuka Rangi

We consider the finite-horizon offline reinforcement learning (RL) setting, and are motivated by the challenge of learning the policy at any step h in dynamic programming (DP) algorithms. To learn this, it is sufficient to evaluate the treatment effect of deviating from the behavioral policy at step h after having optimized the policy for all future steps. Since the policy at any step can affect next-state distributions, the related distributional shift challenges can make this problem far more statistically hard than estimating such treatment effects in the stochastic contextual bandit setting. However, the hardness of many real-world RL instances lies between the two regimes. We develop a flexible and general method called selective uncertainty propagation for confidence interval construction that adapts to the hardness of the associated distribution shift challenges. We show benefits of our approach on toy environments and demonstrate the benefits of these techniques for offline policy learning.

Minyoung Kim, Timothy Hospedales

We tackle the general differentiable meta learning problem that is ubiquitous in modern deep learning, including hyperparameter optimization, loss function learning, few-shot learning and more. These problems are often formalized as Bi-Level Optimizations (BLO). We introduce a novel perspective by turning a given BLO problem into a stochastic optimization, where the inner loss function becomes a smooth probability distribution, and the outer loss becomes an expected loss over the inner distribution. To solve this stochastic optimization, we adopt Stochastic Gradient Langevin Dynamics (SGLD) MCMC to sample inner distribution, and propose a recurrent algorithm to compute the MC-estimated hypergradient. Our derivation is similar to forward-mode differentiation, but we introduce a new first-order approximation that makes it feasible for large models without needing to store huge Jacobian matrices. The main benefits are two fold: i) Our stochastic formulation takes into account uncertainty, which makes the method robust to suboptimal inner optimization or non-unique multiple inner minima due to overparametrization; ii) Compared to existing methods that often exhibit unstable behavior and hyperparameter sensitivity in practice, our method leads to considerably more reliable solutions. We demonstrate that the new approach achieves promising results on diverse meta learning problems and easily scales to learning 87M hyper-parameters in the case of Vision Transformers.

Xiang Huang, Hao Peng, Li Sun, Hui Lin, Chunyang Liu, Jiang Cao, Philip S. Yu

Probabilistic embeddings have several advantages over deterministic embeddings as they map each data point to a distribution, which better describes the uncertainty and complexity of data. Many works focus on adjusting the distribution constraint under the Information Bottleneck (IB) principle to enhance representation learning. However, these proposed regularization terms only consider the constraint of each latent variable, omitting the structural information between latent variables. In this paper, we propose a novel structural entropy-guided probabilistic coding model, named SEPC. Specifically, we incorporate the relationship between latent variables into the optimization by proposing a structural entropy regularization loss. Besides, as traditional structural information theory is not well-suited for regression tasks, we propose a probabilistic encoding tree, transferring regression tasks to classification tasks while diminishing the influence of the transformation. Experimental results across 12 natural language understanding tasks, including both classification and regression tasks, demonstrate the superior performance of SEPC compared to other state-of-the-art models in terms of effectiveness, generalization capability, and robustness to label noise.

Chao Huang, Yushu Shi, Waikeung Wong, Chengliang Liu, Wei Wang, Zhihua Wang, Jie Wen

Medical image segmentation provides useful information about the shape and size of organs, which is beneficial for improving diagnosis, analysis, and treatment. Despite traditional deep learning-based models can extract domain-specific knowledge, they face a generalization bottleneck due to the limited embedded knowledge scope. Vision foundation models have been demonstrated to be effective in extracting generalizable knowledge, but they cannot extract domain-specific knowledge without fine-tuning. In this work, we propose a novel multi-view evidential learning-based framework, which can extract both domain-specific and generalizable knowledge from multi-view features by combining the advantages of traditional and vision foundation models. Specifically, a novel multi-view state space model (MV-SSM) is designed to extract task-related knowledge while removing redundant information within multi-view features. The proposed MV-SSM utilizes Mamba, a state space model, to model cross-view contextual dependencies between domain-specific and generalizable features. Additionally, evidential learning is adopted to quantify the segmentation uncertainty of the model for boundary. In special, variational Dirichlet is introduced to characterize the distribution of the result probabilities, parameterized with collected evidence to quantify uncertainty. As a result, the model can reduce the segmentation uncertainties of boundaries by optimizing the parameters of the Dirichlet distribution. Experimental results on three datasets show that our method obtains superior segmentation performance.

Yuya Hikima, Akiko Takeda

In this study, we consider an optimization problem with uncertainty dependent on decision variables, which has recently attracted attention due to its importance in machine learning and pricing applications. In this problem, the gradient of the objective function cannot be obtained explicitly because the decision-dependent distribution is unknown. Therefore, several zeroth-order methods have been proposed, which obtain noisy objective values by sampling and update the iterates. Although these existing methods have theoretical convergence for optimization problems with decision-dependent uncertainty, they require strong assumptions about the function and distribution or exhibit large variances in their gradient estimators. To overcome these issues, we propose two zeroth-order methods under mild assumptions. First, we develop a zeroth-order method with a new one-point gradient estimator including a variance reduction parameter. The proposed method updates the decision variables while adjusting the variance reduction parameter. Second, we develop a zeroth-order method with a two-point gradient estimator. There are situations where only one-point estimators can be used, but if both one-point and two-point estimators are available, it is more practical to use the two-point estimator. As theoretical results, we show the convergence of our methods to stationary points and provide the worst-case iteration and sample complexity analysis. Our simulation experiments with real data on a retail service application show that our methods output solutions with lower objective values than the conventional zeroth-order methods.

Yanna Ding, Zijie Huang, Xiao Shou, Yihang Guo, Yizhou Sun, Jianxi Gao

Learning curve extrapolation predicts neural network performance from early training epochs and has been applied to accelerate AutoML, facilitating hyperparameter tuning and neural architecture search. However, existing methods typically model the evolution of learning curves in isolation, neglecting the impact of neural network (NN) architectures, which influence the loss landscape and learning trajectories. In this work, we explore whether incorporating neural network architecture improves learning curve modeling and how to effectively integrate this architectural information. Motivated by the dynamical system view of optimization, we propose a novel architecture-aware neural differential equation model to forecast learning curves continuously. We empirically demonstrate its ability to capture the general trend of fluctuating learning curves while quantifying uncertainty through variational parameters. Our model outperforms current state-of-the-art learning curve extrapolation methods and pure time-series modeling approaches for both MLP and CNN-based learning curves. Additionally, we explore the applicability of our method in Neural Architecture Search scenarios, such as training configuration ranking.

Asaf Cassel, Orin Levy, Yishay Mansour

Efficiently trading off exploration and exploitation is one of the key challenges in online Reinforcement Learning (RL). Most works achieve this by carefully estimating the model uncertainty and following the so-called optimistic model. Inspired by practical ensemble methods, in this work we propose a simple and novel batch ensemble scheme that provably achieves near-optimal regret for stochastic Multi-Armed Bandits (MAB). Crucially, our algorithm has just a single parameter, namely the number of batches, and its value does not depend on distributional properties such as the scale and variance of the losses. We complement our theoretical results by demonstrating the effectiveness of our algorithm on synthetic benchmarks.

Cheng Wu, Liang Su, Chaokun Wang, Shaoyun Shi, Ziqian Zhang, Ziyang Liu, Wang Peng, Wenjin Wu, Peng Jiang

Recommender systems are increasingly prevalent to provide personalized suggestions and enhance user satisfaction. Typical recommendation models encode users and items as embeddings, and generate recommendations by assessing the similarity between these embeddings. Despite their effectiveness, these embedding-based models struggle with modeling user uncertainty and capturing diverse user interests using a single fixed user embedding. Recent studies have begun to explore a user-distribution paradigm to learn distributions for users. However, this approach employs a single distribution per user, which fails to effectively delineate semantic boundaries, resulting in sub-optimal recommendations. To this end, we propose GCDR, a Guided Conditional Diffusion Recommender model, to learn multiple distributions for each user in this paper. Specifically, GCDR addresses two major challenges: 1) learning disentangled distributions, and 2) learning personalized distributions. GCDR captures inter-user and intra-user distribution properties through conditional and guided diffusion, respectively. It maintains user-specific embeddings to encode long-term interests for conditional diffusion, while for guided diffusion, it incorporates short-term interests encoded from recent interactions with category preferences. To align the diffusion model with the recommendation task, we train GCDR with three loss functions, included the user loss, the recommendation loss and the diffusion loss. Extensive experiments on four real-world datasets show that GCDR is able to learn effective user distributions and is superior to thirteen state-of-the-art baseline methods.

Yuchen Sun, Qianqian Xu, Zitai Wang, Zhiyong Yang, Junwei He

Multi-label Out-Of-Distribution (OOD) detection aims to discriminate the OOD samples from the multi-label In-Distribution (ID) ones. Compared with its multiclass counterpart, it is crucial to model the joint information among classes. To this end, JointEnergy, which is a representative multi-label OOD inference criterion, summarizes the logits of all the classes. However, we find that JointEnergy can produce an imbalance problem in OOD detection, especially when the model lacks enough discrimination ability. Specifically, we find that the samples only related to minority classes tend to be classified as OOD samples due to the ambiguous energy decision boundary. Besides, imbalanced multi-label learning methods, originally designed for ID ones, would not make sense for OOD detection scenarios, even producing a serious negative transfer effect. In this paper, we resort to auxiliary outlier exposure (OE) and propose an unknown-aware multi-label learning framework to reshape the uncertainty energy space layout. In this framework, the energy score is separately optimized for tail ID samples and unknown samples, and the energy distribution gap between them is expanded, such that the tail ID samples can have a significantly larger energy score than the OOD ones. What's more, a simple yet effective measure is designed to select more informative OE datasets. Finally, comprehensive experimental results on multiple multi-label and OOD datasets reveal the effectiveness of the proposed method.

Zesong Yang, Ru Zhang, Jiale Shi, Zixiang Ai, Boming Zhao, Hujun Bao, Luwei Yang, Zhaopeng Cui

Neural surface representation has demonstrated remarkable success in the areas of novel view synthesis and 3D reconstruction. However, assessing the geometric quality of 3D reconstructions in the absence of ground truth mesh remains a significant challenge, due to its rendering-based optimization process and entangled learning of appearance and geometry with photometric losses. In this paper, we present a novel framework, i.e, GURecon, which establishes a geometric uncertainty field for the neural surface based on geometric consistency. Different from existing methods that rely on rendering-based measurement, GURecon models a continuous 3D uncertainty field for the reconstructed surface, and is learned by an online distillation approach without introducing real geometric information for supervision. Moreover, in order to mitigate the interference of illumination on geometric consistency, a decoupled field is learned and exploited to finetune the uncertainty field. Experiments on various datasets demonstrate the superiority of GURecon in modeling 3D geometric uncertainty, as well as its plug-and-play extension to various neural surface representations and improvement on downstream tasks such as incremental reconstruction.