Score-function policy gradients have delivered strong results in game-playing, robotics and language-model fine-tuning. Yet its high-variance often undermines training stability. On the other hand, pathwise policy gradients alleviate the training variance, but are reliable only when driven by an accurate action-conditioned value function which is notoriously hard to train without relying on past off-policy data. In this paper, we discuss how to construct a value-gradient driven, on-policy algorithm that allow training Q-value models purely from on-policy data, unlocking the possibility of using pathwise policy updates in the context of on-policy learning. We show how to balance stochastic policies for exploration with constrained policy updates for stable training, and evaluate important architectural components that facilitate accurate value function learning. Building on these insights, we propose Relative Entropy Pathwise Policy Optimization (REPPO), an efficient on-policy algorithm that combines the sample-efficiency of pathwise policy gradients with the simplicity and minimal memory footprint of standard on-policy learning. We demonstrate that REPPO provides strong empirical performance at decreased sample requirements, wall-clock time, memory footprint as well as high hyperparameter robustness in a set of experiments on two standard GPU-parallelized benchmarks.
The resurgence of convolutional neural networks (CNNs) in visual recognition tasks, exemplified by ConvNeXt, has demonstrated their capability to rival transformer-based architectures through advanced training methodologies and ViT-inspired design principles. However, both CNNs and transformers exhibit a simplicity bias, favoring straightforward features over complex structural representations. Furthermore, modern CNNs often integrate MLP-like blocks akin to those in transformers, but these blocks suffer from significant information redundancies, necessitating high expansion ratios to sustain competitive performance. To address these limitations, we propose SpaRTAN, a lightweight architectural design that enhances spatial and channel-wise information processing. SpaRTAN employs kernels with varying receptive fields, controlled by kernel size and dilation factor, to capture discriminative multi-order spatial features effectively. A wave-based channel aggregation module further modulates and reinforces pixel interactions, mitigating channel-wise redundancies. Combining the two modules, the proposed network can efficiently gather and dynamically contextualize discriminative features. Experimental results in ImageNet and COCO demonstrate that SpaRTAN achieves remarkable parameter efficiency while maintaining competitive performance. In particular, on the ImageNet-1k benchmark, SpaRTAN achieves 77. 7% accuracy with only 3.8M parameters and approximately 1.0 GFLOPs, demonstrating its ability to deliver strong performance through an efficient design. On the COCO benchmark, it achieves 50.0% AP, surpassing the previous benchmark by 1.2% with only 21.5M parameters. The code is publicly available at [https://github.com/henry-pay/SpaRTAN].
Scaling reinforcement learning (RL) workloads often requires distributing environment simulation across compute clusters. Existing frameworks entangle simulation, learning logic, and orchestration into monolithic systems, limiting modularity and reusability. We present ClusterEnv, a lightweight, learner-agnostic interface for distributed environment execution that mirrors the Gymnasium API. ClusterEnv introduces the DETACH pattern, which decouples simulation from training by offloading reset() and step() operations to remote workers while keeping learning centralized. To address policy staleness in distributed execution, we propose Adaptive Actor Policy Synchronization (AAPS), a divergence-triggered update mechanism that reduces synchronization overhead without sacrificing performance. ClusterEnv integrates cleanly into existing RL pipelines, supports both on-policy and off-policy methods, and requires minimal code changes. Experiments on discrete control tasks demonstrate that AAPS achieves high sample efficiency with significantly fewer weight updates. Source code is available at https://github.com/rodlaf/ClusterEnv.
Cross-view localization, the task of estimating a camera's 3-degrees-of-freedom (3-DoF) pose by aligning ground-level images with satellite images, is crucial for large-scale outdoor applications like autonomous navigation and augmented reality. Existing methods often rely on fully supervised learning, which requires costly ground-truth pose annotations. In this work, we propose GeoDistill, a Geometry guided weakly supervised self distillation framework that uses teacher-student learning with Field-of-View (FoV)-based masking to enhance local feature learning for robust cross-view localization. In GeoDistill, the teacher model localizes a panoramic image, while the student model predicts locations from a limited FoV counterpart created by FoV-based masking. By aligning the student's predictions with those of the teacher, the student focuses on key features like lane lines and ignores textureless regions, such as roads. This results in more accurate predictions and reduced uncertainty, regardless of whether the query images are panoramas or limited FoV images. Our experiments show that GeoDistill significantly improves localization performance across different frameworks. Additionally, we introduce a novel orientation estimation network that predicts relative orientation without requiring precise planar position ground truth. GeoDistill provides a scalable and efficient solution for real-world cross-view localization challenges. Code and model can be found at https://github.com/tongshw/GeoDistill.
Protein language models have emerged as powerful tools for sequence generation, offering substantial advantages in functional optimization and denovo design. However, these models also present significant risks of generating harmful protein sequences, such as those that enhance viral transmissibility or evade immune responses. These concerns underscore critical biosafety and ethical challenges. To address these issues, we propose a Knowledge-guided Preference Optimization (KPO) framework that integrates prior knowledge via a Protein Safety Knowledge Graph. This framework utilizes an efficient graph pruning strategy to identify preferred sequences and employs reinforcement learning to minimize the risk of generating harmful proteins. Experimental results demonstrate that KPO effectively reduces the likelihood of producing hazardous sequences while maintaining high functionality, offering a robust safety assurance framework for applying generative models in biotechnology.
Spatial understanding has been a challenging task for existing Multi-modal Large Language Models~(MLLMs). Previous methods leverage large-scale MLLM finetuning to enhance MLLM's spatial understanding ability. In this paper, we present a data-efficient approach. We propose a LLM agent system with strong and advanced spatial reasoning ability, which can be used to solve the challenging spatial question answering task in complex indoor warehouse scenarios. Our system integrates multiple tools that allow the LLM agent to conduct spatial reasoning and API tools interaction to answer the given complicated spatial question. Extensive evaluations on the 2025 AI City Challenge Physical AI Spatial Intelligence Warehouse dataset demonstrate that our system achieves high accuracy and efficiency in tasks such as object retrieval, counting, and distance estimation. The code is available at: https://github.com/hsiangwei0903/SpatialAgent
Spacecraft deployed in outer space are routinely subjected to various forms of damage due to exposure to hazardous environments. In addition, there are significant risks to the subsequent process of in-space repairs through human extravehicular activity or robotic manipulation, incurring substantial operational costs. Recent developments in image segmentation could enable the development of reliable and cost-effective autonomous inspection systems. While these models often require large amounts of training data to achieve satisfactory results, publicly available annotated spacecraft segmentation data are very scarce. Here, we present a new dataset of nearly 64k annotated spacecraft images that was created using real spacecraft models, superimposed on a mixture of real and synthetic backgrounds generated using NASA's TTALOS pipeline. To mimic camera distortions and noise in real-world image acquisition, we also added different types of noise and distortion to the images. Finally, we finetuned YOLOv8 and YOLOv11 segmentation models to generate performance benchmarks for the dataset under well-defined hardware and inference time constraints to mimic real-world image segmentation challenges for real-time onboard applications in space on NASA's inspector spacecraft. The resulting models, when tested under these constraints, achieved a Dice score of 0.92, Hausdorff distance of 0.69, and an inference time of about 0.5 second. The dataset and models for performance benchmark are available at https://github.com/RiceD2KLab/SWiM.
Vision Language Action (VLA) models represent a transformative shift in robotics, with the aim of unifying visual perception, natural language understanding, and embodied control within a single learning framework. This review presents a comprehensive and forward-looking synthesis of the VLA paradigm, with a particular emphasis on robotic manipulation and instruction-driven autonomy. We comprehensively analyze 102 VLA models, 26 foundational datasets, and 12 simulation platforms that collectively shape the development and evaluation of VLAs models. These models are categorized into key architectural paradigms, each reflecting distinct strategies for integrating vision, language, and control in robotic systems. Foundational datasets are evaluated using a novel criterion based on task complexity, variety of modalities, and dataset scale, allowing a comparative analysis of their suitability for generalist policy learning. We introduce a two-dimensional characterization framework that organizes these datasets based on semantic richness and multimodal alignment, showing underexplored regions in the current data landscape. Simulation environments are evaluated for their effectiveness in generating large-scale data, as well as their ability to facilitate transfer from simulation to real-world settings and the variety of supported tasks. Using both academic and industrial contributions, we recognize ongoing challenges and outline strategic directions such as scalable pretraining protocols, modular architectural design, and robust multimodal alignment strategies. This review serves as both a technical reference and a conceptual roadmap for advancing embodiment and robotic control, providing insights that span from dataset generation to real world deployment of generalist robotic agents.
Language Model (LM) pipelines can dynamically refine their outputs against programmatic constraints. However, their effectiveness collapses when faced with competing soft constraints, leading to inefficient backtracking loops where satisfying one constraint violates another. We introduce Meta Self-Refining, a framework that equips LM pipelines with a meta-corrective layer to repair these competitions at runtime/inference-time. Our approach monitors the pipeline's execution history to detect oscillatory failures. Upon detection, it invokes a meta-repairer LM that analyzes the holistic state of the backtracking attempts and synthesizes a strategic instruction to balance the competing requirements. This self-repair instruction guides the original LM out of a failing refining loop towards a successful output. Our results show Meta Self-Refining can successfully repair these loops, leading to more efficient LM programs.
Modern Artificial Intelligence (AI) increasingly relies on multi-agent architectures that blend visual and language understanding. Yet, a pressing challenge remains: How can we trust these agents especially in zero-shot settings with no fine-tuning? We introduce a novel modular Agentic AI visual classification framework that integrates generalist multimodal agents with a non-visual reasoning orchestrator and a Retrieval-Augmented Generation (RAG) module. Applied to apple leaf disease diagnosis, we benchmark three configurations: (I) zero-shot with confidence-based orchestration, (II) fine-tuned agents with improved performance, and (III) trust-calibrated orchestration enhanced by CLIP-based image retrieval and re-evaluation loops. Using confidence calibration metrics (ECE, OCR, CCC), the orchestrator modulates trust across agents. Our results demonstrate a 77.94\% accuracy improvement in the zero-shot setting using trust-aware orchestration and RAG, achieving 85.63\% overall. GPT-4o showed better calibration, while Qwen-2.5-VL displayed overconfidence. Furthermore, image-RAG grounded predictions with visually similar cases, enabling correction of agent overconfidence via iterative re-evaluation. The proposed system separates perception (vision agents) from meta-reasoning (orchestrator), enabling scalable and interpretable multi-agent AI. This blueprint is extensible to diagnostics, biology, and other trust-critical domains. All models, prompts, results, and system components including the complete software source code are openly released to support reproducibility, transparency, and community benchmarking at Github: https://github.com/Applied-AI-Research-Lab/Orchestrator-Agent-Trust
World models (WMs) demonstrate strong capabilities in prediction, generation, and planning tasks. Existing WMs primarily focus on unstructured data and cannot leverage the ubiquitous structured data, often represented as graphs, in the digital world. While multiple graph foundation models have been proposed, they focus on graph learning tasks and cannot extend to diverse multi-modal data and interdisciplinary tasks. To address these challenges, we propose the Graph World Model (GWM), a world model that supports both unstructured and graph-structured states with multi-modal information and represents diverse tasks as actions. The core of a GWM is a generic message-passing algorithm to aggregate structured information, either over a unified multi-modal token space by converting multi-modal data into text (GWM-T) or a unified multi-modal embedding space by modality-specific encoders (GWM-E). Notably, GWM introduces action nodes to support diverse tasks, where action nodes are linked to other nodes via direct reference or similarity computation. Extensive experiments on six tasks from diverse domains, including multi-modal generation and matching, recommendation, graph prediction, multi-agent, retrieval-augmented generation, and planning and optimization, show that the same GWM outperforms or matches domain-specific baselines' performance, benefits from multi-hop structures, and demonstrates strong zero-shot/few-shot capabilities on unseen new tasks. Our code for GWM is released at https://github.com/ulab-uiuc/GWM.
Federated Learning has emerged as a leading paradigm for decentralized, privacy-preserving learning, particularly relevant in the era of interconnected edge devices equipped with sensors. However, the practical implementation of Federated Learning faces three primary challenges: the need for human involvement in costly data labelling processes for target adaptation, covariate shift in client device data collection due to environmental factors affecting sensors, leading to discrepancies between source and target samples, and the impracticality of continuous or regular model updates in resource-constrained environments due to limited data transmission capabilities and technical constraints on channel availability and energy efficiency. To tackle these issues, we expand upon an efficient and scalable Federated Learning framework tailored for real-world client adaptation in industrial settings. This framework leverages a pre-trained source model comprising a deep backbone, an adaptation module, and a classifier running on a powerful server. By freezing the backbone and classifier during client adaptation on resource-constrained devices, we allow the domain adaptive linear layer to handle target domain adaptation, thus minimizing overall computational overhead. Furthermore, this setup, designated as FedAcross+, is extended to encompass the processing of streaming data, thereby rendering the solution suitable for non-stationary environments. Extensive experimental results demonstrate the effectiveness of FedAcross+ in achieving competitive adaptation on low-end client devices with limited target samples, successfully addressing the challenge of domain shift. Moreover, our framework accommodates sporadic model updates within resource-constrained environments, ensuring practical and seamless deployment.
We propose a multi-camera multi-target (MCMT) tracking framework that ensures consistent global identity assignment across views using trajectory and appearance cues. The pipeline starts with BoT-SORT-based single-camera tracking, followed by an initial glance phase to initialize global IDs via trajectory-feature matching. In later frames, new tracklets are matched to existing global identities through a prioritized global matching strategy. New global IDs are only introduced when no sufficiently similar trajectory or feature match is found. 3D positions are estimated using depth maps and calibration for spatial validation.
Large language models (LLMs) typically require fine-tuning for domain-specific tasks, and LoRA offers a computationally efficient approach by training low-rank adapters. LoRA is also communication-efficient for federated LLMs when multiple users collaboratively fine-tune a global LLM model without sharing their proprietary raw data. However, even the transmission of local adapters between a server and clients risks serious privacy leakage. Applying differential privacy (DP) to federated LoRA encounters a dilemma: adding noise to both adapters amplifies synthetic noise on the model, while fixing one adapter impairs the learnability of fine-tuning. In this paper, we propose FedASK (Differentially Private Federated Low Rank Adaptation with Double Sketching) , a novel federated LoRA framework to enable effective updating of both low-rank adapters with robust differential privacy. Inspired by randomized SVD, our key idea is a two-stage sketching pipeline. This pipeline first aggregates carefully sketched, privacy-preserving local updates, and then reconstructs the global matrices on the server to facilitate effective updating of both adapters. We theoretically prove FedASK's differential privacy guarantee and its exact aggregation property. Comprehensive experiments demonstrate that FedASK consistently outperforms baseline methods across a variety of privacy settings and data distributions.
The remarkable representational power of Vision Transformers (ViTs) remains underutilized in few-shot image classification. In this work, we introduce ViT-ProtoNet, which integrates a ViT-Small backbone into the Prototypical Network framework. By averaging class conditional token embeddings from a handful of support examples, ViT-ProtoNet constructs robust prototypes that generalize to novel categories under 5-shot settings. We conduct an extensive empirical evaluation on four standard benchmarks: Mini-ImageNet, FC100, CUB-200, and CIFAR-FS, including overlapped support variants to assess robustness. Across all splits, ViT-ProtoNet consistently outperforms CNN-based prototypical counterparts, achieving up to a 3.2\% improvement in 5-shot accuracy and demonstrating superior feature separability in latent space. Furthermore, it outperforms or is competitive with transformer-based competitors using a more lightweight backbone. Comprehensive ablations examine the impact of transformer depth, patch size, and fine-tuning strategy. To foster reproducibility, we release code and pretrained weights. Our results establish ViT-ProtoNet as a powerful, flexible approach for few-shot classification and set a new baseline for transformer-based meta-learners.
Digital Twins (DT) have the potential to transform traffic management and operations by creating dynamic, virtual representations of transportation systems that sense conditions, analyze operations, and support decision-making. A key component for DT of the transportation system is dynamic roadway geometry sensing. However, existing approaches often rely on static maps or costly sensors, limiting scalability and adaptability. Additionally, large-scale DTs that collect and analyze data from multiple sources face challenges in privacy, communication, and computational efficiency. To address these challenges, we introduce Geo-ORBIT (Geometrical Operational Roadway Blueprint with Integrated Twin), a unified framework that combines real-time lane detection, DT synchronization, and federated meta-learning. At the core of Geo-ORBIT is GeoLane, a lightweight lane detection model that learns lane geometries from vehicle trajectory data using roadside cameras. We extend this model through Meta-GeoLane, which learns to personalize detection parameters for local entities, and FedMeta-GeoLane, a federated learning strategy that ensures scalable and privacy-preserving adaptation across roadside deployments. Our system is integrated with CARLA and SUMO to create a high-fidelity DT that renders highway scenarios and captures traffic flows in real-time. Extensive experiments across diverse urban scenes show that FedMeta-GeoLane consistently outperforms baseline and meta-learning approaches, achieving lower geometric error and stronger generalization to unseen locations while drastically reducing communication overhead. This work lays the foundation for flexible, context-aware infrastructure modeling in DTs. The framework is publicly available at https://github.com/raynbowy23/FedMeta-GeoLane.git.
Rare events such as state transitions are difficult to observe directly with molecular dynamics simulations due to long timescales. Enhanced sampling techniques overcome this by introducing biases along carefully chosen low-dimensional features, known as collective variables (CVs), which capture the slow degrees of freedom. Machine learning approaches (MLCVs) have automated CV discovery, but existing methods typically focus on discriminating meta-stable states without fully encoding the detailed dynamics essential for accurate sampling. We propose TLC, a framework that learns CVs directly from time-lagged conditions of a generative model. Instead of modeling the static Boltzmann distribution, TLC models a time-lagged conditional distribution yielding CVs to capture the slow dynamic behavior. We validate TLC on the Alanine Dipeptide system using two CV-based enhanced sampling tasks: (i) steered molecular dynamics (SMD) and (ii) on-the-fly probability enhanced sampling (OPES), demonstrating equal or superior performance compared to existing MLCV methods in both transition path sampling and state discrimination.
Perceiving and reconstructing 4D spatial-temporal geometry from videos is a fundamental yet challenging computer vision task. To facilitate interactive and real-time applications, we propose a streaming 4D visual geometry transformer that shares a similar philosophy with autoregressive large language models. We explore a simple and efficient design and employ a causal transformer architecture to process the input sequence in an online manner. We use temporal causal attention and cache the historical keys and values as implicit memory to enable efficient streaming long-term 4D reconstruction. This design can handle real-time 4D reconstruction by incrementally integrating historical information while maintaining high-quality spatial consistency. For efficient training, we propose to distill knowledge from the dense bidirectional visual geometry grounded transformer (VGGT) to our causal model. For inference, our model supports the migration of optimized efficient attention operator (e.g., FlashAttention) from the field of large language models. Extensive experiments on various 4D geometry perception benchmarks demonstrate that our model increases the inference speed in online scenarios while maintaining competitive performance, paving the way for scalable and interactive 4D vision systems. Code is available at: https://github.com/wzzheng/StreamVGGT.
In text-to-image generation, producing a series of consistent contents that preserve the same identity is highly valuable for real-world applications. Although a few works have explored training-free methods to enhance the consistency of generated subjects, we observe that they suffer from the following problems. First, they fail to maintain consistent background details, which limits their applicability. Furthermore, when the foreground character undergoes large motion variations, inconsistencies in identity and clothing details become evident. To address these problems, we propose CharaConsist, which employs point-tracking attention and adaptive token merge along with decoupled control of the foreground and background. CharaConsist enables fine-grained consistency for both foreground and background, supporting the generation of one character in continuous shots within a fixed scene or in discrete shots across different scenes. Moreover, CharaConsist is the first consistent generation method tailored for text-to-image DiT model. Its ability to maintain fine-grained consistency, combined with the larger capacity of latest base model, enables it to produce high-quality visual outputs, broadening its applicability to a wider range of real-world scenarios. The source code has been released at https://github.com/Murray-Wang/CharaConsist
Neural populations exhibit latent dynamical structures that drive time-evolving spiking activities, motivating the search for models that capture both intrinsic network dynamics and external unobserved influences. In this work, we introduce LangevinFlow, a sequential Variational Auto-Encoder where the time evolution of latent variables is governed by the underdamped Langevin equation. Our approach incorporates physical priors -- such as inertia, damping, a learned potential function, and stochastic forces -- to represent both autonomous and non-autonomous processes in neural systems. Crucially, the potential function is parameterized as a network of locally coupled oscillators, biasing the model toward oscillatory and flow-like behaviors observed in biological neural populations. Our model features a recurrent encoder, a one-layer Transformer decoder, and Langevin dynamics in the latent space. Empirically, our method outperforms state-of-the-art baselines on synthetic neural populations generated by a Lorenz attractor, closely matching ground-truth firing rates. On the Neural Latents Benchmark (NLB), the model achieves superior held-out neuron likelihoods (bits per spike) and forward prediction accuracy across four challenging datasets. It also matches or surpasses alternative methods in decoding behavioral metrics such as hand velocity. Overall, this work introduces a flexible, physics-inspired, high-performing framework for modeling complex neural population dynamics and their unobserved influences.