论文检索

Bram Grooten, Patrick MacAlpine, Kaushik Subramanian, Peter Stone, Peter R. Wurman

Generalization to unseen environments is a significant challenge in the field of robotics and control. In this work, we focus on contextual reinforcement learning, where agents act within environments with varying contexts, such as self-driving cars or quadrupedal robots that need to operate in different terrains or weather conditions than they were trained for. We tackle the critical task of generalizing to out-of-distribution (OOD) settings, without access to explicit context information at test time. Recent work has addressed this problem by training a context encoder and a history adaptation module in separate stages. While promising, this two-phase approach is cumbersome to implement and train. We simplify the methodology and introduce SPARC: single-phase adaptation for robust control. We test SPARC on varying contexts within the high-fidelity racing simulator Gran Turismo 7 and wind-perturbed MuJoCo environments, and find that it achieves reliable and robust OOD generalization.

Alonso Granados, Jason Pacheco

Model-free deep reinforcement learning (RL) algorithms have achieved tremendous success on a range of challenging tasks. However, safety concerns remain when these methods are deployed on real-world applications, necessitating risk-aware agents. A common utility for learning such risk-aware agents is the entropic risk measure, but current policy gradient methods optimizing this measure must perform high-variance and numerically unstable updates. As a result, existing risk-sensitive model-free approaches are limited to simple tasks and tabular settings. In this paper, we provide a comprehensive theoretical justification for policy gradient methods on the entropic risk measure, including on- and off-policy gradient theorems for the stochastic and deterministic policy settings. Motivated by theory, we propose risk-sensitive exponential actor-critic (rsEAC), an off-policy model-free approach that incorporates novel procedures to avoid the explicit representation of exponential value functions and their gradients, and optimizes its policy w.r.t. the entropic risk measure. In this way, we show that rsEAC produces more numerically stable updates compared to existing approaches and reliably learns risk-sensitive policies in challenging risky variants of continuous tasks in MuJoCo.

Cameron Gordon, Yiping Ji, Hemanth Saratchandran, Paul Albert, Simon Lucey

Resource-constrained weight deployment is a task of immense practical importance. Recently, there has been interest in the specific task of Delta Compression, where parties each hold a common base model and only communicate compressed weight updates. However, popular parameter efficient updates such as Low Rank Adaptation (LoRA) face inherent representation limitations - which are especially pronounced when combined with aggressive quantization. To overcome this, we build on recent work that improves LoRA representation capacity by using fixed-frequency sinusoidal functions to increase stable rank without adding additional parameters. We extend this to the quantized setting and present the first theoretical analysis showing how stable rank evolves under quantization. From this, we introduce SineLoRA∆, a principled and effective method for delta compression that improves the expressivity of quantized low-rank adapters by applying a sinusoidal activation. We validate SineLoRA∆ across a diverse variety of domains - including language modeling, vision-language tasks, and text-to-image generation - achieving up to 66% memory reduction with similar performance. We additionally provide a novel application of the canonical Bjøntegaard Delta metric to consistently compare adapter compression changes across the rate-distortion curve.

Alex W. Goodall, Edwin Hamel-De Le Court, Francesco Belardinelli

Many reinforcement learning algorithms, particularly those that rely on return estimates for policy improvement, can suffer from poor sample efficiency and training instability due to high-variance return estimates. In this paper we leverage new results from off-policy evaluation; it has recently been shown that well-designed behaviour policies can be used to collect off-policy data for provably lower variance return estimates. This result is surprising as it means collecting data on-policy is not variance optimal. We extend this key insight to the online reinforcement learning setting, where both policy evaluation and improvement are interleaved to learn optimal policies. Off-policy RL has been well studied (e.g., IMPALA), with correct and truncated importance weighted samples for de-biasing and managing variance appropriately. Generally these approaches are concerned with reconciling data collected from multiple workers in parallel, while the policy is updated asynchronously, mismatch between the workers and policy is corrected in a mathematically sound way. Here we consider only one worker - the behaviour policy, which is used to collect data for policy improvement, with provably lower variance return estimates. In our experiments we extend two policy-gradient methods with this regime, demonstrating better sample efficiency and performance over a diverse set of environments.

Mingrong Gong, Chaoqi Chen, Luyao Tang, Yuxi Wang, Sergio Escalera

4D point cloud segmentation is crucial for autonomous driving with continuous LiDAR streams. While test-time adaptation (TTA) is the standard approach for handling dynamic environments, current methods suffer from catastrophic error accumulation due to over-reliance on pseudo-labels. Active learning could provide reliable annotations for critical samples, but combining it with TTA faces severe challenges: realtime processing requirements and expensive 3D labeling costs. In this paper, we propose ATTA-4DSeg, the first framework to achieve efficient active test-time adaptation for 4D point cloud segmentation under extreme budget constraints. Our key insight is a self-reinforcing loop: oracle annotations refine adaptation prototypes, which then guide the selection of subsequent high-value samples from regions with severe distribution shifts, maximizing each annotation’s impact. Specifically, we propose three key innovations: (1) dual-prototype comparison that precisely localizes distribution shift boundaries to narrow annotation scope, (2) Class-Inverse Budget Allocation (CIBA) ensuring balanced adaptation across all categories, coupled with hybrid uncertainty scoring combining voxel-level geometry and point-wise variance for optimal sample selection, and (3) a refinement strategy leveraging sparse oracle annotations to improve predictions on unlabeled points, maximizing annotation utility. Extensive experiments show ATTA-4DSeg improves mIoU by 18.87%, 19.92%, and 3.6% on three domain adaptation benchmarks using only 1% annotation budget. Our method operates 2.28× faster than state-of-the-art methods. Remarkably, our approach reaches 90% of fully-supervised performance using only 5% annotation budget.

Lluis Gomez

Membership Inference Attacks (MIAs) test whether a model has memorized training data, and are a key tool for auditing privacy risks in machine learning. Recent papers report near-perfect MIA success against large vision-language models such as CLIP, but almost all evaluations train on one web-scale corpus (e.g. LAION-400M) and treat samples from a different corpus (e.g. COCO or CC12M) as non-members - thereby turning the task into out-of-distribution (OOD) detection rather than true membership testing, introducing spurious signals unrelated to true memorization. We revisit the problem with a distribution-matched benchmark built from the CommonPool-L corpus of DataComp. A ViT-B/16 CLIP trained on 400M pairs is accompanied by two 26-shard, i.i.d. splits that serve as member and non-member sets, sharing the exact same acquisition and preprocessing pipeline. Under this strictly in-distribution setting, every published MIA baseline collapses to chance (~51% AUC). To explain this collapse, we derive a scaling-law upper bound for similarity-based attacks showing that the expected member vs. non-member similarity gap decays as O(T/N) for contrastive learning with T epochs over N samples. Empirically, as we vary the training set size while holding all hyper-parameters fixed, the gap follows the predicted linear trend in log–log space, and Cosine Similarity Attack AUC drops from 94% to 51%. Finally, we propose a simple, white-box, gradient-based MIA that outperforms prior attacks for CLIP without relying on OOD cues. We release code, checkpoints, and data to foster comprehensive and reproducible privacy research on multimodal CLIP-like foundation models.

Joseph Giovanelli, Giuseppe Pisano, Roberta Calegari

AI systems can perpetuate and amplify existing biases and discrimination, prompting academic efforts to develop mitigation techniques. Despite progress, real-world deployments often expose limitations in current methods and tools--- overlooking preprocessing, adopting poor evaluation protocols, and failing to integrate domain knowledge. These gaps hinder the effectiveness and reproducibility of fairness solutions. AutoML has emerged as a promising approach to optimize AI pipelines and provide an evaluation framework. However, challenges persist, especially around: intersectionality support, explainability, and stakeholder engagement, which are crucial for fairness and human-centric AI development. We introduce HAMLET4Fairness, integrating AutoML with human-centered approaches grounded in logic and argumentation. This enhances interactivity and transparency in AI pipeline optimization while supporting intersectional fairness. HAMLET4Fairness leverages multi-objective optimization and bounds the search space by user-defined constraints, adapting the CRISP-DM methodology for co-design and collaborative problem solving. We validate HAMLET4Fairness through the well-known case studies in the literature and provide insights into how preprocessing choices affect fairness.

Toan D. Gian, Mohammad Abdi, Nathaniel D. Bastian, Francesco Restuccia

HARL enables agents to execute cooperative tasks by adopting agent-specific policies. Most of existing HARL methods use individual policy neural networks to ensure monotonic improvement, which leads to substantial computational overhead. The proposed SDE-HARL overcomes this limitation by decomposing each agent's policy neural network into a lightweight local neural network and a global neural network executed at an edge server. Each local neural network generates and sends a compressed latent representation to the edge server, which aggregates the representations and produces agent-specific inferences. As such, SDE-HARL allows to significantly save computing and networking resources while preserving agent-specific behavior. A key feature of SDE-HARL is grouping agents with similar roles via a role-aware mechanism and share partial parameters in their global networks, while an identity-aware mechanism is introduced to promote behavioral diversity among agents within the same group. We prototyped SDE-HARL on an experimental testbed composed of a Jetson Nano and Raspberry PI to measure latency and network resource consumption. We evaluated SDE-HARL's performance on several benchmark datasets, including Google Research Football and StarCraft II. Experimental results show that SDE-HARL reaches up to 90% win rate while reducing latency, energy consumption, and networking overhead respectively by 2x, 2.5x, and 5x compared to existing work.

Debamita Ghosh, George K. Atia, Yue Wang

Reinforcement learning (RL) faces significant challenges in real-world deployments due to the sim-to-real gap, where policies trained in simulators often underperform in practice due to mismatches between training and deployment conditions. Distributionally robust RL addresses this issue by optimizing worst-case performance over an uncertainty set of environments and providing an optimized lower bound on deployment performance. However, existing studies typically assume access to either a generative model or offline datasets with broad coverage of the deployment environment—assumptions that limit their practicality in unknown environments without prior knowledge. In this work, we study the more realistic and challenging setting of online distributionally robust RL, where the agent interacts only with a single unknown training environment while aiming to optimize its worst-case performance. We focus on general f-divergence-based uncertainty sets, including chi-squared and KL divergence balls, and propose a computationally efficient algorithm with sublinear regret guarantees under minimal assumptions. Furthermore, we establish a minimax lower bound on regret of online learning, demonstrating the near-optimality of our approach. Extensive experiments across diverse environments further confirm the robustness and efficiency of our algorithm, validating our theoretical findings.

Aoyun Geng, Chunyan Cui, Yunyun Su, Zhenjie Luo, Feifei Cui, Zilong Zhang

With the rapid advance of spatial multi-omics technologies, it has become possible to simultaneously profile transcripts, proteins and chromatin states at their native spatial coordinates, thereby uncovering molecular architecture that transcends any single-omics perspective. However, the resulting data matrices are often highly sparse and suffer from unstable dimensionality. Graph-based neural methods capture only local neighborhood information, whereas conventional Transformers, although capable of modelling long-range dependencies, incur prohibitive computational costs on such data. To overcome these limitations, we propose TLAGC—a Taylor-Linear-Attention-Guided Graph Convolutional framework that couples a Taylor-expanded linear attention (TLA) mechanism with graph convolutional networks. By eliminating the soft-max operation and linking the LocalGCN via residual connections, TLA preserves local structural information while enabling the integration of global and local contexts, thereby alleviating ineffective information propagation between spatially distant yet transcriptionally similar regions. Theoretical analysis confirms that TLA indeed reduces computational complexity, and extensive experiments on multiple spatial multi-omics benchmarks demonstrate that TLAGC consistently outperforms state-of-the-art baselines in delineating spatial domains.

Chandan Gautam, Lew Choon Hean, Ankit Das, Xiaoli Li, Savitha Ramasamy

Generalized Category Discovery (GCD) aims to classify labeled instances from known categories while discovering novel categories from unlabeled data. Despite recent progress in GCD for computer vision, existing GCD approaches largely rely on static final-step representations (in the visual domain), overlooking the temporally evolving nature of time-series data. In this paper, we introduce TGCD, the first framework specifically designed for GCD in time-series data. TGCD leverages both the dynamics of latent representations and the heterogeneity of predictions across multiple temporal segments to disover unknown (i.e., novel) categories, based on a pre-trained time-series foundation model. We propose a unified learning objective for TGCD that integrates the following three components: (i) a Stochastic Temporal Segment Dropout (STeSD) objective that regularizes the model by selectively penalizing high-entropy segments to encourage confident predictions on uncertain regions of the time-series, and (ii) a Known–Unknown Temporal Discriminability (KUTD) objective that promotes representational separation between known and unknown categories within unlabeled data and (iii) a margin-aware classification objective to improve generalization. Empirical evaluation on six multivariate time-series data sets demonstrates that the TGCD substantially outperforms existing GCD methods, particularly in discovering unknown categories. We further conduct ablation studies to highlight the individual contributions of each component. Additionally, we provide the first comprehensive benchmarking of recent GCD approaches on time-series data, revealing the limitations of naive transfer and underscoring the benefits of temporal modeling.

Ziqi Gao

Understanding when a pre-trained model generalizes well to a new task remains a key challenge in transfer learning. Classical theories bound target risk using divergences such as total variation, MMD, or Wasserstein distance, yet tasks with similar divergences often show very different transfer performance. We propose a structural framework that explains transferability through two factors: the Feature Overlap Rate (FOR), measuring how much target representation lies in the source-induced subspace, and the Effective Task Complexity (ETC), quantifying the entropy of latent subtasks. We derive a PAC-Bayesian bound where target risk depends on FOR and ETC, and show that larger models attenuate their negative effects. Experiments on six GLUE transfer pairs estimate FOR and ETC from encoder representations and compare them to classical divergences. Results show that FOR and ETC together explain over 80% of transfer risk variance, while divergences fail to do so. Our findings provide a geometry-aware perspective for diagnosing and guiding transfer learning.

Yongbiao Gao, Xiangcheng Sun, Chao Tan, Chunyu Hu, Guohua Lv

Label Distribution Learning (LDL) is a groundbreaking paradigm for addressing the task with label ambiguity. Subjectivity in annotating label description degrees often leads to imbalanced label distribution. Existing approaches either adopt representation alignment or decoupling strategies to solve the imbalanced label distribution learning (ILDL). However, representation alignment-based methods overlook the issue of gradient vanishing for non-dominant branches within imbalanced label distributions, while decoupling-based approaches fail to achieve adaptive weight optimization. To address these issues, we propose Adaptive Momentum and Exponential Moving Average weighted modeling (AMEMA). AMEMA combines EMA-based loss weighting with momentum allocation to mitigate gradient attenuation in non-dominant label learning and adaptively balance the optimization signals between dominant and non-dominant branches. It computes and updates Kullback-Leibler divergence losses for each branch using EMA, and applies different initial momenta to facilitate branch-specific optimization dynamics. Dynamic weighting coefficients, derived from EMA-smoothed losses, allow the model to adjust its learning direction adaptively and improve the learning of non-dominant labels. Extensive experiments on benchmark datasets show that AMEMA consistently outperforms state-of-the-art ILDL methods across various evaluation metrics.

Yilin Gao, Shuguang Dou, Junzhou Li, Zhiheng Yu, Yin Li, Dongsheng Jiang, Shugong Xu

Recent advances in diffusion models have achieved remarkable success in isolated computer vision tasks such as text-to-image generation, depth estimation, and optical flow. However, these models are often restricted by a ``single-task-single-model'' paradigm, severely limiting their generalizability and scalability in multi-task scenarios. Motivated by the cross-domain generalization ability of large language models, we propose a universal visual perception framework based on flow matching that can generate diverse visual representations across multiple tasks. Our approach formulates the process as a universal flow-matching problem from image patch tokens to task-specific representations rather than an independent generation or regression problem. By leveraging a strong self-supervised foundation model as the anchor and introducing a multi-scale, circular task embedding mechanism, our method learns a universal velocity field to bridge the gap between heterogeneous tasks, supporting efficient and flexible representation transfer. Extensive experiments on classification, detection, segmentation, depth estimation, and image-text retrieval demonstrate that our model achieves competitive performance in both zero-shot and fine-tuned settings, outperforming prior generalist and several specialist models. Ablation studies further validate the robustness, scalability, and generalization of our framework. Our work marks a significant step towards general-purpose visual perception, providing a solid foundation for future research in universal vision modeling.

Xiyuan Gao, Bing Cao, Baoquan Gong, Pengfei Zhu

Multimodal learning frequently faces two coupled challenges: modality imbalance, where dominant modalities suppress others during training, and modality conflict, where opposing gradient directions hinder optimization. Existing methods typically address these issues in isolation, yet they are intrinsically correlated and most fundamentally reflected in the gradient space—severe imbalance may obscure conflicts, while suppressing conflict may homogenize features and worsen imbalance, affecting fusion performance. To jointly address this coupled challenge, we propose Reconcile Gradient Modulation (RGM), a unified framework that adaptively adjusts gradient magnitude and direction for harmony multimodal learning. The core of RGM is SynOrth Grad, which minimizes Dirichlet energy to perform minimal-gradient surgery. It enhances cooperation synergy when modalities are aligned and enforces orthogonality to preserve uniqueness in conflict situations, thus promoting stable and balanced learning. To guide this modulation, we propose Cumulative Gradient Energy (CGE) as a convergence-guaranteed measure of modality-wise progress, and construct a Balance-nonConflict Plane (BCP) for real-time diagnosis and control of training dynamics. Experiments on diverse benchmarks validate our effectiveness and generalizability, consistently outperforming counterparts that are designed to handle multimodal imbalance or conflict independently.

Xiangxiang Gao, Weisheng Xie, Lixin, Xuwei Fang, Chen Hang, Changqun Li, Yuhan Lin, Xiaolong Xu

Large Language Models face fundamental deployment challenges due to the computational demands of auto-regressive token-by-token generation. While speculative decoding has emerged as a promising acceleration technique through its draft-then-verify framework, current implementations suffer from two critical limitations: (1) mutual waiting problem caused by sequential dependencies between draft generation and verification phases, and (2) constrained token acceptance rates where retrieval-based drafting methods under-perform in general domains while models-based drafting approaches show reduced efficacy in knowledge-intensive scenarios. To address these challenges, we propose Talon, a novel parallel inference architecture featuring two key innovations: (1) **a novel asynchronous execution paradigm** that decouples draft generation from verification, effectively eliminating synchronization bottlenecks, and (2) **an adaptive hybrid drafting strategy** that dynamically combines model-based and retrieval-based approaches to improve token acceptance rates across diverse domains. Extensive evaluations across standard benchmarks (MT-Bench, HumanEval, GSM8K, Alpaca, CNN/DM) demonstrate Talon's exceptional performance, achieving 4.04x–6.52x acceleration across multiple model families including Vicuna, Deepseek, and LLaMA series. These results represent a significant advancement over existing speculative decoding methods (EAGLE 1-3, Hydra, Medusa, Lookahead, SPS, and PLD), establishing a new paradigm for speculative decoding.

Wanfu Gao, Zebin He, Jun Gao

Existing feature engineering methods based on large language models (LLMs) have not yet been applied to multi-label learning tasks. They lack the ability to model complex label dependencies and are not specifically adapted to the characteristics of multi-label tasks. To address the above issues, we propose Feature Engineering Automation for Multi-Label Learning (FEAML), an automated feature engineering method for multi-label classification which leverages the code generation capabilities of LLMs. By utilizing metadata and label co-occurrence matrices, LLMs are guided to understand the relationships between data features and task objectives, based on which high-quality features are generated.The newly generated features are evaluated in terms of model accuracy to assess their effectiveness, while Pearson correlation coefficients are used to detect redundancy. FEAML further incorporates the evaluation results as feedback to drive LLMs to continuously optimize code generation in subsequent iterations. By integrating LLMs with a feedback mechanism, FEAML realizes an efficient, interpretable and self-improving feature engineering paradigm. Empirical results on various multi-label datasets demonstrate that our FEAML outperforms other feature engineering methods.

Shengbo Gao, Jinyang Guo, Lixian Su, Yifu Ding, Shiqiao Gu, Aishan Liu, Yuqing Ma, Zhiwang Zhang, Xianglong Liu

Large Language Models (LLMs) hold significant potential for enhancing healthcare applications, yet their deployment is hindered by high computational and memory demands. Model compression techniques offer solutions to reduce these demands, but their impact on medical LLMs remains underexplored. In this paper, we introduce CMedBench, the first comprehensive benchmark for evaluating compressed LLMs in medical contexts. CMedBench assesses five core dimensions: Medical Knowledge Ability, Medical Application Ability, Trustworthiness Maintenance, Compression Cross Combination, and Computational Efficiency. Through extensive empirical studies, we analyze the trade-offs between model efficiency and clinical performance across diverse models, datasets, and compression strategies. Our findings highlight critical limitations in current evaluation practices and provide a robust framework for aligning compression strategies with medical requirements. CMedBench serves as a vital resource for researchers and practitioners, guiding the development of efficient, trustworthy, and clinically effective LLMs for healthcare applications.

Ruixuan Gao, Qijun Zhao, Keren Fu

Non-Exemplar Class Incremental Learning (NECIL) strives to preserve classification performance in an evolving data stream without revisiting old-class exemplars. Current methods mitigate catastrophic forgetting by replaying and augmenting historical prototypes as surrogates for old classes. However, they treat prototypes as holistic representations for global-level augmentations, which overlook dimensional semantic disparity and old-new class relationships, failing to maintain old-class discriminability and adaptability to the evolving feature space. To address this challenge, we propose Dimensionally-Allocated Prototype Refinement (DiAPR), a granular framework that progressively refines prototypes to exhibit class separability in the new feature space through three modules. Specifically, Distribution-aware Pairing (DAP) captures old-new class semantic consistency to guide Granular Semantic Allocation (GSA) in dimension-wise conflation, while Cross-Dimensional Transition (CDT) enhances cross-dimensional dependencies. The resulting prototypes sharpen classifier decision boundaries. Moreover, CDT inherently enables softened feature alignment, thereby yielding a more compatible feature space. Extensive experiments demonstrate DiAPR’s superiority, with improvements over SOTA by 2.35%, 0.70%, 0.96% on three CIFAR-100 settings, 1.03%, 0.54%, 0.40% on Tiny-ImageNet, and 0.60% on ImageNet-Subset.

Qingqing Gao, Tengfei Liu, Xiaoyan Li, Xiaodan Zhang, Zhongfan Sun, Boyue Wang, Baocai Yin, Zhaohui Liu

Radiology report generation from longitudinal medical data is critical for assessing disease progression and automating diagnostic workflows. While recent methods incorporate longitudinal information, they primarily rely on multimodal feature fusion, with limited capacity for explicit disease evolution modeling and temporal reasoning. To address this, we propose MARE, an end-to-end framework that formulates longitudinal radiology report generation as a multimodal analogical reasoning task. Inspired by the Abduction–Mapping–Induction paradigm, MARE models latent relational structures underlying disease evolution by aligning lesion-level visual features across time and mapping them to the textual domain for temporally coherent and clinically meaningful report generation. To mitigate the spatial misalignment caused by patient positioning or imaging variation, we introduce an Adaptive Region Alignment (ARA) module for robust temporal correspondence. Additionally, we design Dual Evolution Consistency (DEC) losses to regularize analogical reasoning by enforcing temporal coherence in both visual and textual evolution paths. Extensive experiments on the Longitudinal-MIMIC dataset demonstrate that MARE significantly outperforms state-of-the-art baselines across both natural language generation and clinical effectiveness metrics, highlighting the value of structured analogical reasoning for disease evolution-aware report generation.