Searcharxiv⌕ Search

arXiv subjects

Dan Lu

Publications and source records attributed to Dan Lu.

At least 19 recordsLinked to original sources

Learning When to Collaborate: Selective Multi-Agent Medical Reasoning via Uncertainty-Aware Routing

Multi-agent systems have emerged as a promising paradigm for multimodal medical reasoning, but existing methods rely on two assumptions that hinder practical deployment: (i) always-active dense collaboration among multiple agents, and (ii) dependence on large cloud-based models that raise privacy and efficiency concerns. In this paper, we challenge both assumptions and propose SMART-Med, a fully local multi-agent method that treats collaboration as a learnable decision rather than a default behavior. Our key insight is that not every medical query requires costly multi-agent deliberation: many can be reliably solved by a single well-trained agent, while only uncertain cases benefit from expert collaboration. Based on this insight, SMART-Med introduces an uncertainty-driven routing mechanism that adaptively decides when to answer directly and when to recruit a small subset of expert agents, effectively converting dense multi-agent reasoning into selective, query-adaptive coordination. To make this routing reliable, we curate a new multimodal medical dataset MedVR-3K with high-quality reasoning traces and a difficulty hierarchy, and then fine-tune local agents using GRPO with two complementary rewards: a certainty-aware accuracy reward and a vision-language reward model for correctness and reasoning quality respectively. Experiments on four medical visual question answering benchmarks show that SMART-Med achieves competitive performance compared to existing MAS approaches while substantially reducing inference cost (reducing inference time by more than 3.8x and token usage by 72%). Moreover, after fine-tuning on MedVR-3K, SMART-Med yields an average improvement of 6.7% across Slake, PathVQA, VQA-RAD, and PMC-VQA.

cs.AI↗

ST-Bench: A Spatial-Temporal Benchmark for Multi-Agent System Generation on Scientific Research Tasks

The rapid progress of LLM-based multi-agent systems (MAS) has shown that they largely outperform single agents on coding, math, and QA tasks, where executable tests provide a binary success signal. Whether this advantage transfers to real scientific data analysis remains untested. We introduce ST-Bench, a benchmark designed to answer two questions: whether MAS outperform single agents on complex scientific data analysis tasks, and if so, by how much and at what additional cost. ST-Bench contains 100 data science tasks adapted from published Earth science studies across hydrology, agriculture, and wetland methane research, expanded into 2,067 queries grounded in additional published studies and validated by domain experts. Using ST-Bench, we evaluate five recent MAS generation methods under two training protocols, against single-agent baselines on the same GPT-5 backbone. Nine of the ten MAS configurations exceed the cheapest single-agent baseline, with the strongest reaching nearly three times its composite score. This gain is primarily attributable to coverage: trained workflows produce realistic numerical metrics on a larger fraction of queries, while the quality of those metrics, conditional on producing realistic output, is comparable to that of the single-agent baseline. The strongest configuration requires approximately four times the single-agent inference time, whereas a more economical workflow captures the majority of the benefit at less than twice the cost. MAS specialization confers measurable benefit on scientific data analysis, but the benefit is conditional rather than universal.

cs.AI↗

ZeroDiff: Zero-Shot Time Series Reconstruction via Informed-Prior Diffusion

Time series modeling increasingly demands high-quality supervision, yet target observations remain scarce - exogenous inputs are broadly available, but target measurements are often unavailable due to cost, infrastructure, or accessibility constraints. Can models trained on observed locations reconstruct target time series where measurements have never been collected? We term this zero-shot time series reconstruction. A naive approach - directly mapping exogenous inputs to targets - can yield predictions at unobserved locations, but without target signals, such models fail to capture the intrinsic dynamics of the target variable, producing overly smooth outputs that underestimate extremes. This reveals systematic errors that call for explicit modeling and calibration. We propose ZeroDiff, which constructs an informed prior from exogenous variables alone, then learns to calibrate reconstruction errors through diffusion - training on observed locations and generalizing to unobserved ones. Experiments across diverse real-world datasets demonstrate significant improvements over existing approaches. Our code is available at https://github.com/YingdaFan/ZeroDiff-ICML2026.

cs.LG↗

Global Attention with Linear Complexity for Exascale Generative Data Assimilation in Earth System Prediction

Accurate Earth system prediction requires state inference from incomplete observations, but conventional two-stage data assimilation (DA) is computationally prohibitive because repeated PDE-based ensemble forecasts, observation updates, and intermediate data movement limit ensemble size at high resolution. We introduce STORM, a one-stage generative AI framework that reformulates DA as diffusion-based Bayesian posterior sampling, replacing online PDE ensemble forecasts with scalable AI inference. It further combines a spatiotemporal transformer with a global-attention algorithm that reduces complexity from quadratic to linear through scalable gradient propagation, enabling high-resolution, long-context Earth modeling. STORM scales to 74,400 GPUs on Frontier with 96--99\% strong-scaling efficiency and up to 6 ExaFLOPs sustained BF16 throughput, while enabling 32,768-member ensembles for uncertainty quantification in 34 seconds on 4,096 GPUs. It scales to 20 billion spatiotemporal tokens and 177,000 temporal frames. Hurricane tracking and long-term climate reanalysis demonstrate improved accuracy, including benefits from longer temporal context and recovery of temperature extremes missed by forecast-only predictions.

cs.LG↗

DyFrDet: Towards Accurate Small Object Detection via Dynamic Frequency Suppression with Label Disambiguation

Despite the remarkable progress over the past decades, accurately identifying small objects remains challenging because of their insufficient visual cues. Previous works typically attempt to construct discriminative representation of the small objects. However, the wide range frequency domain noises and label ambiguities have been greatly overlooked, which significantly hinders the accurate localization. To address these issues, we propose a novel small object detection (SOD) detector termed DyFrDet, which is able to precisely localize the small object by dynamically suppressing the background distractions in frequency domain. Specifically, we propose a Dynamic Frequency-aware Feature Pyramid Network (DyFrFPN) to adaptively suppress low-frequency redundancy and excessive high-frequency noises. The DyFrFPN transforms the hierarchical features into frequency domain representation, and introduces a Dynamic Band Predictor (DBP) to preserve the discriminative components for small object identification. Afterwards, we present a novel Label Disambiguation Module (LDM), which leverages probabilistic distributions to explicitly model and alleviate the inherent ambiguity of target labels, yielding efficient improvement in localization precision of the small objects with low-resolution. Extensive experiments demonstrate that DyFrDet achieves state-of-the-art performance across multiple benchmarks, indicating its effectiveness and robustness in various challenging scenarios. Our code is available at https://github.com/ManOfStory/DyFrDet.

cs.CV↗

Conjugate phase-noise cancellation enables submicrometre dual-comb ranging with free-running megahertz-linewidth lasers

Frequency-domain dual-comb ranging combines rapid acquisition with interferometric sensitivity, but high-performance implementations often rely on mutually coherent or actively stabilised comb sources. Free-running sources can reduce this hardware burden, but their phase noise and drift of the optical frequency offset can blur radio-frequency (RF) comb teeth and weaken probe-reference phase correlation. Previous phase-slope implementations have therefore relied on sufficiently resolved RF teeth, within-coherence-length probe-reference paths or explicit digital tracking of these fluctuations. Here we demonstrate a low-cost frequency-domain dual-comb ranging architecture that combines independent free-running distributed-feedback (DFB) lasers with conjugate phase-noise cancellation (CPNC). By forming a self-conjugate signal before the phases of individual RF-comb teeth are extracted, CPNC cancels the common laser phase factor and drifting optical-frequency-offset term while retaining the distance-dependent phase slope. Using electro-optic combs seeded by DFB lasers with linewidths of 12 MHz and 9 MHz, we achieve an Allan deviation of $219~\mathrm{nm}$ at $246~μ\mathrm{s}$ and reduce the single-frame distance standard deviation from $558~μ\mathrm{m}$ to $9.31~μ\mathrm{m}$ with CPNC. Across the tested megahertz-linewidth configurations, CPNC delivered minimum Allan deviations below $250~\mathrm{nm}$. These results show that CPNC enables submicrometre ranging in a low-cost frequency-domain dual-comb architecture with reduced source-stabilisation and phase-management complexity.

physics.optics↗

Emerging Flexible Designs for Geospatial Multimodal Foundation Models

Foundation models are rapidly transforming Earth observation by enabling scalable pretraining across diverse unlabeled geospatial modalities. However, their architectural diversity ranging from encoder-only to encoder-decoder and masked autoencoding paradigms makes it challenging to assess performance trade offs in a consistent manner. In this work, we present an apples-to-apples comparison of leading FM architectures designed for geospatial multimodal reasoning, with a particular focus on flexibility across varied spectral band configurations. We standardize pretraining using identical self supervised learning objectives and training datasets, and evaluate all models under consistent parameterization on the GEOBench benchmark across classification and segmentation tasks. Our results offer new insights into the design trade-offs between model flexibility, modality alignment, and downstream task performance. By highlighting architectural strengths and limitations under controlled conditions, this study provides practical guidance for building next generation geospatial foundation models capable of robust multimodal reasoning.

cs.LG↗

Physical probes expose and alleviate chemical-environment collapse in molecular representations

Nuclear magnetic resonance (NMR) spectroscopy provides an experimental readout of local chemical environments, but its use in molecular representation learning has been constrained by heterogeneous data and incomplete atom-level assignments. Here we construct complementary high-fidelity experimental and computational 13C NMR resources, which reveal a recurrent form of representational collapse: atoms that are equivalent in molecular topology can remain experimentally distinct in their real chemical environments, whereas explicit 3D descriptions are further limited by static conformations in dynamic regimes. To alleviate this bottleneck, we develop CLAIM (Contrastive Learning for Atom-to-molecule Inference of Molecular NMR), a framework that aligns efficient topological molecular inputs with atom-resolved NMR observables. Through hierarchical chemical priors and cross-level contrastive learning, CLAIM restores lost chemical resolution and markedly improves atom-level molecule-spectrum retrieval. CLAIM remains robust in flexible and tautomeric systems for 13C NMR prediction, improves stereoisomer discrimination without explicit 3D modelling, and transfers to broader molecular property tasks including ADMET prediction and fluorescence estimation. These results establish physically grounded spectral alignment as an effective strategy for alleviating chemical-environment collapse and for guiding experimentally grounded molecular representation learning.

physics.chem-ph↗

Towards Adaptive Humanoid Control via Multi-Behavior Distillation and Reinforced Fine-Tuning

Humanoid robots are promising to learn a diverse set of human-like locomotion behaviors, including standing up, walking, running, and jumping. However, existing methods predominantly require training independent policies for each skill, yielding behavior-specific controllers that exhibit limited generalization and brittle performance when deployed on irregular terrains and in diverse situations. To address this challenge, we propose Adaptive Humanoid Control (AHC) that adopts a two-stage framework to learn an adaptive humanoid locomotion controller across different skills and terrains. Specifically, we first train several primary locomotion policies and perform a multi-behavior distillation process to obtain a basic multi-behavior controller, facilitating adaptive behavior switching based on the environment. Then, we perform reinforced fine-tuning by collecting online feedback in performing adaptive behaviors on more diverse terrains, enhancing terrain adaptability for the controller. We conduct experiments in both simulation and real-world experiments in Unitree G1 robots. The results show that our method exhibits strong adaptability across various situations and terrains. Project website: https://ahc-humanoid.github.io.

cs.RO↗

HydroDCM: Hydrological Domain-Conditioned Modulation for Cross-Reservoir Inflow Prediction

Deep learning models have shown promise in reservoir inflow prediction, yet their performance often deteriorates when applied to different reservoirs due to distributional differences, referred to as the domain shift problem. Domain generalization (DG) solutions aim to address this issue by extracting domain-invariant representations that mitigate errors in unseen domains. However, in hydrological settings, each reservoir exhibits unique inflow patterns, while some metadata beyond observations like spatial information exerts indirect but significant influence. This mismatch limits the applicability of conventional DG techniques to many-domain hydrological systems. To overcome these challenges, we propose HydroDCM, a scalable DG framework for cross-reservoir inflow forecasting. Spatial metadata of reservoirs is used to construct pseudo-domain labels that guide adversarial learning of invariant temporal features. During inference, HydroDCM adapts these features through light-weight conditioning layers informed by the target reservoir's metadata, reconciling DG's invariance with location-specific adaptation. Experiment results on 30 real-world reservoirs in the Upper Colorado River Basin demonstrate that our method substantially outperforms state-of-the-art DG baselines under many-domain conditions and remains computationally efficient.

cs.LG↗

Adaptive Graph Learning with Transformer for Multi-Reservoir Inflow Prediction

Reservoir inflow prediction is crucial for water resource management, yet existing approaches mainly focus on single-reservoir models that ignore spatial dependencies among interconnected reservoirs. We introduce AdaTrip as an adaptive, time-varying graph learning framework for multi-reservoir inflow forecasting. AdaTrip constructs dynamic graphs where reservoirs are nodes with directed edges reflecting hydrological connections, employing attention mechanisms to automatically identify crucial spatial and temporal dependencies. Evaluation on thirty reservoirs in the Upper Colorado River Basin demonstrates superiority over existing baselines, with improved performance for reservoirs with limited records through parameter sharing. Additionally, AdaTrip provides interpretable attention maps at edge and time-step levels, offering insights into hydrological controls to support operational decision-making. Our code is available at https://github.com/humphreyhuu/AdaTrip.

cs.LG↗

ORBIT-2: Scaling Exascale Vision Foundation Models for Weather and Climate Downscaling

Sparse observations and coarse-resolution climate models limit effective regional decision-making, underscoring the need for robust downscaling. However, existing AI methods struggle with generalization across variables and geographies and are constrained by the quadratic complexity of Vision Transformer (ViT) self-attention. We introduce ORBIT-2, a scalable foundation model for global, hyper-resolution climate downscaling. ORBIT-2 incorporates two key innovations: (1) Residual Slim ViT (Reslim), a lightweight architecture with residual learning and Bayesian regularization for efficient, robust prediction; and (2) TILES, a tile-wise sequence scaling algorithm that reduces self-attention complexity from quadratic to linear, enabling long-sequence processing and massive parallelism. ORBIT-2 scales to 10 billion parameters across 65,536 GPUs, achieving up to 4.1 exaFLOPS sustained throughput and 74--98% strong scaling efficiency. It supports downscaling to 0.9 km global resolution and processes sequences up to 4.2 billion tokens. On 7 km resolution benchmarks, ORBIT-2 achieves high accuracy with $R^2$ scores in the range of 0.98--0.99 against observational data.

cs.LG↗

Decoding Memories: An Efficient Pipeline for Self-Consistency Hallucination Detection

Large language models (LLMs) have demonstrated impressive performance in both research and real-world applications, but they still struggle with hallucination. Existing hallucination detection methods often perform poorly on sentence-level generation or rely heavily on domain-specific knowledge. While self-consistency approaches help address these limitations, they incur high computational costs due to repeated generation. In this paper, we conduct the first study on identifying redundancy in self-consistency methods, manifested as shared prefix tokens across generations, and observe that non-exact-answer tokens contribute minimally to the semantic content. Based on these insights, we propose a novel Decoding Memory Pipeline (DMP) that accelerates generation through selective inference and annealed decoding. Being orthogonal to the model, dataset, decoding strategy, and self-consistency baseline, our DMP consistently improves the efficiency of multi-response generation and holds promise for extension to alignment and reasoning tasks. Extensive experiments show that our method achieves up to a 3x speedup without sacrificing AUROC performance.

cs.CL↗

Distributed Cross-Channel Hierarchical Aggregation for Foundation Models

Vision-based scientific foundation models hold significant promise for advancing scientific discovery and innovation. This potential stems from their ability to aggregate images from diverse sources such as varying physical groundings or data acquisition systems and to learn spatio-temporal correlations using transformer architectures. However, tokenizing and aggregating images can be compute-intensive, a challenge not fully addressed by current distributed methods. In this work, we introduce the Distributed Cross-Channel Hierarchical Aggregation (D-CHAG) approach designed for datasets with a large number of channels across image modalities. Our method is compatible with any model-parallel strategy and any type of vision transformer architecture, significantly improving computational efficiency. We evaluated D-CHAG on hyperspectral imaging and weather forecasting tasks. When integrated with tensor parallelism and model sharding, our approach achieved up to a 75% reduction in memory usage and more than doubled sustained throughput on up to 1,024 AMD GPUs on the Frontier Supercomputer.

cs.LG↗

Use of LLMs in preparing accessible scientific papers

Making scientific papers accessible may require reprocessing old papers to create output compliant with accessibility standards. An important step there is to convert the visual formatting to the logical one. In this report we describe our attempt at zero shot conversion of arXiv papers. Our results are mixed: while it is possible to do conversion, the reliability is not too good. We discuss alternative approaches to this problem.

cs.DL↗

Diffusion-based supervised learning of generative models for efficient sampling of multimodal distributions

We propose a hybrid generative model for efficient sampling of high-dimensional, multimodal probability distributions for Bayesian inference. Traditional Monte Carlo methods, such as the Metropolis-Hastings and Langevin Monte Carlo sampling methods, are effective for sampling from single-mode distributions in high-dimensional spaces. However, these methods struggle to produce samples with the correct proportions for each mode in multimodal distributions, especially for distributions with well separated modes. To address the challenges posed by multimodality, we adopt a divide-and-conquer strategy. We start by minimizing the energy function with initial guesses uniformly distributed within the prior domain to identify all the modes of the energy function. Then, we train a classifier to segment the domain corresponding to each mode. After the domain decomposition, we train a diffusion-model-assisted generative model for each identified mode within its support. Once each mode is characterized, we employ bridge sampling to estimate the normalizing constant, allowing us to directly adjust the ratios between the modes. Our numerical examples demonstrate that the proposed framework can effectively handle multimodal distributions with varying mode shapes in up to 100 dimensions. An application to Bayesian inverse problem for partial differential equations is also provided.

stat.ML↗

Predicting the critical behavior of complex dynamic systems via learning the governing mechanisms

Critical points separate distinct dynamical regimes of complex systems, often delimiting functional or macroscopic phases in which the system operates. However, the long-term prediction of critical regimes and behaviors is challenging given the narrow set of parameters from which they emerge. Here, we propose a framework to learn the rules that govern the dynamic processes of a system. The learned governing rules further refine and guide the representative learning of neural networks from a series of dynamic graphs. This combination enables knowledge-based prediction for the critical behaviors of dynamical networked systems. We evaluate the performance of our framework in predicting two typical critical behaviors in spreading dynamics on various synthetic and real-world networks. Our results show that governing rules can be learned effectively and significantly improve prediction accuracy. Our framework demonstrates a scenario for facilitating the representability of deep neural networks through learning the underlying mechanism, which aims to steer applications for predicting complex behavior that learnable physical rules can drive.

physics.soc-ph↗

Recommendations for Comprehensive and Independent Evaluation of Machine Learning-Based Earth System Models

Machine learning (ML) is a revolutionary technology with demonstrable applications across multiple disciplines. Within the Earth science community, ML has been most visible for weather forecasting, producing forecasts that rival modern physics-based models. Given the importance of deepening our understanding and improving predictions of the Earth system on all time scales, efforts are now underway to develop forecasting models into Earth-system models (ESMs), capable of representing all components of the coupled Earth system (or their aggregated behavior) and their response to external changes. Modeling the Earth system is a much more difficult problem than weather forecasting, not least because the model must represent the alternate (e.g., future) coupled states of the system for which there are no historical observations. Given that the physical principles that enable predictions about the response of the Earth system are often not explicitly coded in these ML-based models, demonstrating the credibility of ML-based ESMs thus requires us to build evidence of their consistency with the physical system. To this end, this paper puts forward five recommendations to enhance comprehensive, standardized, and independent evaluation of ML-based ESMs to strengthen their credibility and promote their wider use.

cs.LG↗