We introduce missingness-MDPs (miss-MDPs), a novel subclass of partially observable Markov decision processes (POMDPs) that incorporates the theory of missing data. A miss-MDP is a POMDP whose observation function is a missingness function, specifying the probability that individual state features are missing (i.e., unobserved) at a time step. The literature distinguishes three canonical missingness types: (1) missing completely at random (MCAR), (2) missing at random (MAR), and (3) missing not at random (MNAR). The problem is to compute near-optimal policies for a miss-MDP with an unknown missingness function, given a dataset of action-observation histories. Achieving such optimality guarantees for policies requires learning the missingness function from data, which is infeasible for general POMDPs. To overcome this challenge, we exploit the structural properties of different missingness types to derive probably approximately correct (PAC) algorithms for learning the missingness function. These algorithms yield an approximate but fully specified miss-MDP that we solve using off-the-shelf planning methods. We prove that, with high probability, the resulting policies are ε-optimal in the true miss-MDP. Empirical results confirm the theory and demonstrate superior performance of our approach over two model-free methods.
论文检索
输入标题、作者或关键词，从 990 篇学术成果中精准定位
Attributing model behavior to training data is an evolving research field. A common benchmark is data removal, which involves eliminating data instances with either low or high values, then assessing a model's performance trained on the modified dataset. Many existing studies leverage Shapley-based data values for this task. In this paper, we demonstrate that these data values are not optimally suited for pruning low-value data when only a limited amount of data remains. To address this limitation, we introduce the Constraint-Data-Value-Maximization (CDVM) approach, which effectively utilizes data attributions for pruning in low-data scenarios. By casting pruning as a constrained optimization that both maximizes total influence and penalizes excessive per-test contributions, CDVM delivers robust performance when only a small fraction of the data is retained. On the OpenDataVal benchmark, CDVM shows strong performance and competitive runtime.
Preference alignment is a critical step in making Large Language Models (LLMs) useful and aligned with (human) preferences. Existing approaches such as Reinforcement Learning from Human Feedback or Direct Preference Optimization typically require curated data and expensive optimization over billions of parameters, and eventually lead to persistent task-specific models. In this work, we introduce Preference alignment of Large Language Models via Residual Steering (PaLRS), a training-free method that exploits preference signals encoded in the residual streams of LLMs. From as few as one hundred preference pairs, PaLRS extracts lightweight, plug-and-play steering vectors that can be applied at inference time to push models toward preferred behaviors. We evaluate PaLRS on various small-to-medium-scale open-source LLMs, showing that PaLRS-aligned models achieve consistent gains on mathematical reasoning and code generation benchmarks while preserving baseline general-purpose performance. Moreover, when compared to models aligned with DPO and SimPO, they perform better with great time savings. Our findings highlight that PaLRS offers an effective, much more efficient and flexible alternative to standard preference optimization pipelines, offering a training-free, plug-and-play mechanism for alignment with minimal data. Extended version with Suppl. Mat. is available at https://doi.org/10.48550/arXiv.2509.23982.
Cross-view Referring Expression Comprehension (REC) requires a model to accurately localize objects in a spatial representation of one perspective based on descriptions generated from a different perspective. Essentially, it demands maintaining referential consistency under perspective shifts. This problem is prevalent in engineering application scenarios (e.g., Geographic Information Systems (GIS), urban planning, etc), where descriptions are typically generated from a top-down or planning view, while targets must be identified in a spatial representation distinct from the source. However, existing methods often assume shared perspectives or rely on an explicit observer to interpret observer-centric spatial relations (e.g., "left/right" of the observer). Consequently, they struggle to handle cross-view grounding cases commonly arising in engineering application scenarios, where descriptions are defined by the source-view representation itself and do not involve any observing subject. To this end, we propose a novel Subject-Agnostic Cross-View REC task and construct a controlled simulation environment to systematically study referential consistency under perspective changes. Methodologically, we propose a Reward-Driven Consistency Learning framework. By imposing mutually constraining reward signals on the grounding and localization results between the source and target views, we guide the model to maintain referential consistency across different view representations. Experimental results show that our method significantly outperforms baselines in multiple cross-view settings, exhibiting superior generalization and robustness under perspective shifts.
Large EEG Models (LEMs) are drawing increasing attention in EEG, as large-scale pretraining yields transferable representations that improve generalization. As EEG research moves to real-world deployment, objectives and paradigms diversify, yielding increasingly heterogeneous and unevenly scaled datasets across institutions. This evolution induces distribution drift and scale imbalance, making static pretrained LEMs brittle and requiring adaptation to newly collected data. Under the constraints of the high cost of retraining and the privacy sensitivity of newly collected data, federated learning offers a practical route to update LEMs via decentralized parameter updates. Yet standard federated fine-tuning is unstable for pretrained LEMs, suffering backbone drift, misaligned updates under non-IID heterogeneity, and scale imbalance. We propose FedGLoRA: we freeze the backbone and communicate only LoRA adapters via a dual-branch design, with a shared global branch and a private on-device branch. FedGLoRA aggregates global updates on a Grassmann subspace with a Grassmann-proximal constraint, and uses bidirectional distillation to mitigate imbalance. Experiments on motor imagery and emotion recognition show consistent gains and improved stability over strong baselines.
In recent years, theoretical understanding has rapidly advanced regarding how popular multi-objective evolutionary algorithms (MOEAs) can optimize many-objective problems. However, the benefits of using crossover in many-objective optimization are theoretically not understood, except for specifically designed benchmark functions tuned to particular crossover operators, and still lag significantly behind its practical use. In this paper, we build upon this line of research and present a theoretical runtime analysis of the widely used NSGA-III algorithm on the classical m-objective m-OneJumpZeroJump function (m-OJZJ for short). Our results demonstrate that NSGA-III with crossover optimizes m-OJZJ asymptotically faster than NSGA-III without crossover for any number m of objectives for huge parameter regimes. We complement our analysis by providing a lower runtime bound on 4-OJZJ when crossover is turned off.
World models (WM) learn predictive representations from high-dimensional sensory input. However, their internal knowledge is implicit and often inadequate for systematic reasoning about object structure. Qualitative Spatial Reasoning (QSR) models provide explicit human-readable object descriptions under transformations. This paper shows that these two forms of modeling can be combined by injecting a QSR model into a WM to improve the WM reasoning abilities. Using a cube rotation setting inspired by the cube comparison test used in mental rotation tasks as a case study, we use a QSR model that captures the geometric relationships and transformation rules describing how features move across faces under discrete rotations. We translate this QSR model into differentiable constraints using Logic Tensor Networks (LTNs) to guide gradient based learning. These constraints are injected into the WM loss function to regularize its latent dynamics to satisfy the spatial constraints of rotation. The results show that under an extreme data scarce regime (10%), baseline model's performance collapses to near zero, whereas our approach achieves above 90% better performance relative to baseline. Our method consistently maintains high scores across all data regimes (~80%), for both in-distribution and out-of-distribution observations. This paper proposes a novel framework to infuse QSR knowledge inside WM to make learned latent dynamics logical. Code is available at: https:// github.com/ajayrao80/LtnDreamer/
Machine learning (ML)-based malware detectors are widely deployed but remain vulnerable to adversarial attacks. However, under hard-label black-box access, existing adversarial attacks on Windows Portable Executable (PE) malware are often query-inefficient and incur large file-size inflation. A common paradigm is to predefine a set of semantics-preserving atomic perturbations and search for evasive combinations under only binary feedback. Among these atomic perturbations, (i) those with highly combinatorial search spaces are difficult to explore effectively under hard-label feedback, leaving their potential untapped, and (ii) those relying on transplanting benign fragments are often laden with evasion-irrelevant bytes and exhibit highly variable adversarial utility. We propose Gradient-seeded Reinforcement Learning And Stealthy Pruning (GRASP), a three-stage framework that tackles these challenges. First, we decouple perturbations with highly combinatorial search spaces from the query-based search and instead apply a gradient-seeded warm-up that uses Gumbel-Softmax relaxation to enable gradient-based updates over the discrete space. This yields a strong warm start that improves evasion and reduces queries in later stages. Second, Reinforcement Learning (RL)-based refinement is accelerated by a perturbation library that filters, caches, and reuses compact high-utility patterns, reducing wasted queries on low-utility benign fragments. Third, a perturbation minimization stage removes redundant bytes while preserving evasion, reducing size inflation and feeding compact patterns back to the library. Experiments show that GRASP outperforms baselines, achieving higher attack success with fewer queries and smaller file-size inflation. We additionally demonstrate its practical effectiveness against commercial Antivirus engines.
Retrieval-Augmented Generation (RAG) is sensitive to the vast hyperparameters of the retriever and generator, yet optimizing them using given queries remains a challenging task due to the complex hyperparameter interactions and expensive evaluation costs. Existing algorithms are ineffective and slow in convergence, since they often treat RAG as a monolithic black box or only optimize partial hyperparameters. In this paper, we propose CDS4RAG, a framework that optimizes the full hyperparameters of RAG using given queries via a new formulation of cyclic dual-sequential problem. CDS4RAG is special in the sense that it distinguishes the hyperparameters of the retriever and generator, optimizing them in turn and in a cyclic manner. Such a paradigm allows us to design fine-grained within-cycle budget provision and expedite the optimization via cross-cycle seeding when optimizing the generator. Importantly, CDS4RAG is an algorithm-agnostic framework that can be paired with diverse general algorithms. Through experiments on four common benchmarks and two backbone LLMs, we reveal that CDS4RAG considerably boosts the vanilla algorithms in 21/24 cases while significantly outperforming state-of-the-art algorithms in all cases with up to 1.54x improvements of generation quality and better speedup.
In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks with considerable performance gains, yet its effectiveness is highly sensitive to the choice of demonstrations. Most existing selection methods rely on heuristic or proxy signals, such as similarity, diversity, or uncertainty, and select demonstrations independently, which may misalign with downstream performance and overlook set-level composition effects. Therefore, we propose Performance-Driven Demonstration Selection (PDDS), which directly aligns demonstration selection with ICL performance. PDDS formulates selection as predicting the target LLM’s downstream task performance for a given query–in-context pair, replacing proxy heuristics with a performance-aware objective. Unlike prior approaches that rank demonstrations independently before composing the prompt, PDDS evaluates the entire in-context set, capturing inter-demonstration interactions and composition effects. PDDS trains an end-to-end scorer using supervision from the target LLM’s actual task outcomes. At inference time, it selects high-scoring in-context sets without additional target LLM calls or validation-time feedback. Across 6 NLP tasks, 8 datasets, and 8 target LLMs ranging from 1B to 70B parameters, PDDS achieves state-of-the-art results and generalizes well across LLMs and datasets. PDDS remains effective with as few as 30 training instances and can be integrated as a plug-and-play component to enhance existing ICL methods.
With the expanding use of artificial intelligence generated images (AGIs) in scenarios such as gaming, art, and film production, evaluating their quality is essential to ensure their practical utility. To guarantee effective quality measurement, both content and distortion must be considered. However, existing image quality assessment (IQA) methods fail to sufficiently combine the two aspects in their assessments. Meanwhile, the AGIs annotated with credible content and distortion labels are lack, thus, employing cross-domain methods to utilize typically public image datasets is meaningful. Based on the above conclusions, this paper proposed a cross-domain AI-generated IQA via content-distortion awareness (CDAQA). Firstly, the large language model has been utilized to obtain content and distortion expressions of images. Secondly, a dual-stream contrastive language-image pre-training module (DCLIP) has been designed to gain both quality-aware features in images and the corresponding expressions that related to content and distortion, respectively. Thirdly, a graph knowledge transferring module (GKT) that constructs graphs via inter-domain content relevance has been proposed. GKT can update distortion information in the source domain according to the target domain representations. Finally, the cross-domain framework has been designed to update the existing IQA model for AGIs. And, the contrastive loss of DCLIP, feature update loss of GKT, and quality prediction loss are coordinated to jointly update different modules, enabling sufficient fusion of target domain knowledge. Experimental results demonstrate that CDAQA achieves higher accuracy and stability in cross-domain AGIs tasks.
Unifying Perspectives: Plausible Counterfactual Explanations on Global, Group-wise, and Local Levels
The growing complexity of AI systems has intensified the need for transparency through Explainable AI (XAI). Counterfactual explanations (CFs) offer actionable "what-if" scenarios on three levels: Local CFs providing instance-specific insights, Global CFs addressing broader trends, and Group-wise CFs (GWCFs) striking a balance and revealing patterns within cohesive groups. Despite the availability of methods for each granularity level, the field lacks a unified method that integrates these complementary approaches. We address this limitation by proposing a gradient-based optimization method for differentiable models that generates Local, Global, and Group-wise Counterfactual Explanations in a unified manner. We especially enhance GWCF generation by combining instance grouping and counterfactual generation into a single efficient process, replacing traditional two-step methods. Moreover, to ensure trustworthiness, we innovatively introduce the integration of plausibility criteria into the GWCF domain, making explanations both valid and realistic. Our results demonstrate the method's effectiveness in balancing validity, proximity, and plausibility while optimizing group granularity, with practical utility validated through practical use cases.
Decentralized Multi Agent Reinforcement Learning (MARL) faces a fundamental dilemma in real world deployments: agents must operate under epistemic fragmentation, where local observations are severely occluded, while navigating heterogeneous value landscapes, where sparse, critical events carry disproportionately high stakes. Existing paradigms typically decouple state estimation from policy optimization, guiding perception modules merely to minimize uniform reconstruction error. This leads to a Perception Value Misalignment, where agents squander computational resources reconstructing task irrelevant background noise while failing to resolve uncertainties in high value regions. To bridge this gap, we propose EVA-Gen (Epistemic Value Alignment via Generative Models) that establishes a cybernetic loop between generative perception and value based decision making. We formulate the Value Conditioned Reconstruction Paradigm, establishing that optimal perception under resource constraints is functionally weighted by the gradient of the value function. EVA-Gen couples Backward Flow to steer diffusion toward high stakes manifolds, Collaborative Information Bottleneck to filter communication for value relevant consensus, and Risk Sensitive Rectification to prevent sparse signal dilution, synergistically closing the perception control loop. Empirically, we demonstrate that EVA-Gen achieves superior performance in three value-heterogeneous multi agent environments.
Existing spatial domain identification methods primarily use graph neural networks to model spatial and transcriptional relationships. However, their performance is highly sensitive to noisy affinity graphs. Moreover, graph autoencoders tend to over-constrain latent representations, which limits their ability to capture global variability in spatial multi-omics data. To overcome these issues, we propose a dual-flow latent refinement framework that simultaneously integrates structure-aware and structure-free transformations. Specifically, a graph normalizing flow is employed to enforce relational consistency, while a parallel vanilla normalizing flow preserves global distributional flexibility. Features learned by the two flows are then adaptively fused to obtain a robust and unified latent representations. In addition, we introduce a grouping belief-based affinity refinement strategy to suppress unreliable connections and strengthen confident neighborhood relationships, which provides a more stable structural prior for representation learning. Extensive experiments on multiple spatial multi-omics datasets show that the proposed method consistently outperforms state-of-the-art approaches and achieves more accurate and robust spatial domain identification. The supplementary material is publicly available at https://github.com/LiangSDNULab/GBFlow.
Sleep staging is pivotal for assessing sleep quality and clinical diagnostics. While contactless radio-frequency (RF) sensing offers an unobtrusive alternative to Polysomnography (PSG), existing methods often struggle with generalization across diverse devices and environments due to the scarcity of annotated RF data. To overcome this limitation, we propose XSensorSleep, a cross-sensor domain generalization framework. Uniquely, XSensorSleep is trained exclusively on large-scale respiratory datasets from chest and abdominal belts and is directly transferable to RF-derived respiratory signals. To bridge the domain gap, we partition the training data into multiple pseudo-domains based on data sources and sensor types, forcing the model to extract domain-invariant features from heterogeneous respiratory patterns. Furthermore, we introduce a hierarchical alignment strategy: Epoch-Level Feature Alignment (ELFA) suppresses sensor-specific morphological artifacts, while Spectral Temporal Alignment (STA) captures invariant global sleep architectures. Specifically, STA leverages the rotation-invariance of eigenvalue spectra to align whole-night transition dynamics, ensuring the model prioritizes universal physiological rhythms over device-dependent signal fluctuations. Extensive evaluations across various home and clinical RF datasets demonstrate that XSensorSleep achieves superior generalization, enabling practical, automated, and label-free non-contact sleep staging.
Large Reasoning Models (LRMs) often suffer from overthinking due to redundant verification steps. Existing approaches for mitigating overthinking, such as fast-slow thinking switching and reasoning trajectory compression, fail to make a fine-grained distinction between beneficial and redundant steps within the LRM's reasoning process, and may thus impair reasoning capability in their pursuit of efficiency. To simultaneously improve reasoning efficiency and capability, we propose EvoThink, a framework that reduces redundant verification and encourages the exploration of new reasoning paths. EvoThink comprises two key components: Self-Pruning Training (SPT), an unsupervised method that iteratively prunes redundant reasoning steps and self-trains on the concise trajectories; and Aha-Moment Preference Optimization (AMPO), which, inspired by genetic algorithms, identifies valuable failed reasoning attempts, synthesizes from-wrong-to-right aha-moment data, and optimizes the model to internalize this reasoning pattern. Extensive evaluations across mathematical reasoning and code generation benchmarks demonstrate that EvoThink not only substantially reduces inference-time token usage but also improves the reasoning capability of LRMs.
Large language models (LLMs) dominate both everyday and specialized applications, but their high computational demand, energy consumption, and privacy risks are increasingly critiqued. Small language models (SLMs) mitigate these drawbacks and are gaining momentum in scenarios where full LLM capabilities are not required, such as agents, industrial systems, or edge devices. Nevertheless, a systematic comparison of model capabilities, energy usage, and scaling behavior has not been conducted yet. We evaluate 70+ SLMs from 2023–2025 on five task-specific benchmarks and compare them with two popular LLMs, revealing key trade-offs between energy, performance, and model selection. Our findings challenge common assumptions: First, smaller models are not automatically more efficient, and energy increases do not guarantee performance gains. Second, newer SLMs show clear improvements in performance–energy trade-offs, though the progress begins to plateau. Last, the efficiency landscape forms a clear Pareto frontier: initial energy increases yield substantial gains, but the last percentage points of performance need orders of magnitude more energy. These results highlight diminishing returns of scaling and emphasize the need for informed, task-aware model selection rather than size-driven choices.
Spatial multi-omics technologies jointly profile diverse molecular modalities with spatial context, providing a comprehensive view of cellular heterogeneity and tissue organization. To integrate spatial multi-omics data and identify spatial domains, a wide range of unsupervised methods has been proposed. However, recent approaches rely on single-step fusion, directly aggregating heterogeneous modalities into a shared representation, making embeddings sensitive to modality imbalance and measurement noise that are inherent to spatial multi-omics data, as well as unstable optimization under weak supervision. Here, we propose ILR-SMO (Iterative Latent Refinement for Spatial Multi-Omics), a unified self-supervised framework that reformulates multimodal integration as a stability-aware refinement process over a shared latent representation. Instead of single-step plain fusion, ILR-SMO progressively integrates modality-specific information through sequential updates, allowing information to be injected in a controlled manner. This incremental refinement enables later updates to correct or compensate for earlier deviations, resulting in more stable and robust representation learning under heterogeneous and noisy modalities. ILR-SMO further incorporates reliability-aware modality gating for adaptive modulation of modality contributions, and employs a joint objective to enforce spatial coherence while preventing representation collapse. Extensive experiments on five spatial multi-omics benchmarks demonstrate that ILR-SMO consistently outperforms seven state-of-the-art methods and exhibits strong robustness across diverse settings.
Fair clustering has garnered considerable attention, with various fairness notions proposed to ensure equitable representation across demographic groups. In this paper, we focus on the k-center problem in bounded doubling metrics under two popular fairness requirements: group fairness and data summarization fairness, referred to Group Fair k-Center (Gf-k-Cen) and Data Summarization Fair k-Center (Dsf-k-Cen), respectively. Both fairness notions extend classical clustering formulation by associating each data point with a demographic label. Motivated by recent advances in parameterized approximation results for fair clustering, we investigate whether these problems admit Fixed-Parameter Tractable (FPT) approximation schemes in bounded doubling metrics. The previous algorithms typically neglect the local structural properties induced by fairness constraints itself, which limits their approximation quality. By further leveraging the geometric properties of doubling metrics together with local fairness information, we develop a candidate-based structural method that yields (1+eps)-approximation algorithms with FPT running times for both problems, parameterized by the number of selected centers. To the best of our knowledge, these results constitute the first parameterized approximation schemes for the Gf-k-Cen and Dsf-k-Cen problems in bounded doubling metrics.
Asset pricing and portfolio recommendation are two closely related fundamental tasks in quantitative investment, for which machine learning methods have attracted significant attention in both academia and industry. In particular, nonlinear asset pricing models based on deep learning architectures that learn risk factors and risk exposures (betas) conditioned on high-dimensional asset characteristics have become widely used in the field. Despite their popularity, three challenges remain for portfolio recommendation: (i) their static risk pricing structure constrains predictive performance for expected returns; (ii) their representation learning is typically deterministic or fails to account for the inherent distributional uncertainty in the feature space arising from noisy returns and heterogeneous characteristics; and (iii) the sparse factor structure of asset returns and the diversity of characteristics make it difficult to identify incremental predictive information. To address these issues and bridge the gap to practical applications, we propose a novel multi-task dynamic factor model that jointly performs asset pricing and portfolio recommendation. Specifically, we introduce dual-adversarial trainers into the variational prior-posterior learning framework for Factor and Beta Networks, augmented with probabilistic equivariance regularization and dual adaptive spatio-temporal clustering. These components filter redundant information and enhance the model's ability to adapt to changing market conditions. Extensive experiments on a comprehensive open-source stock market dataset demonstrate that our model achieves strong and robust performance relative to baseline methods in the literature.