arXiv subjects

Yuan Gao

Publications and source records attributed to Yuan Gao.

At least 55 records · Page 3Linked to original sources

arXiv2026

Pailitao-MMSearch: Building Native E-Commerce Multimodal Search Foundation

The evolution of e-commerce has fundamentally transformed how users search for products, shifting from simple text-based keyword queries to complex multimodal interactions that seamlessly combine product images, natural language descriptions, and mixed-intent instructions. However, existing approaches face a critical dilemma: single-modal specialist models, deployed independently for text retrieval, visual search, and voice recognition, operate in isolation and cannot handle cross-modal queries, while general-purpose vision-language models lack the domain-specific knowledge necessary for fine-grained product understanding, user behavior modeling, and commercial intent reasoning. In this work, we present Pailitao-MMSearch, one native e-commerce multimodal search foundation model designed to bridge this gap. Our approach introduces three key innovations: (1)HybSID (Hybrid Semantic ID);(2)a two-stage continual pre-training strategy; and (3)a hybrid reasoning post-training pipeline. Built upon Qwen and deployed on Taobao's Pailitao multimodal search platform, Pailitao-MMSearch achieves substantial improvements in online A/B testing, including up to +13.61\% in Gross Merchandise Volume (GMV) and +8.21\% in transaction volume compared to traditional multi-modal search pipeline, demonstrating the effectiveness of our native e-commerce multimodal search large language models.

cs.AI↗

arXiv2026

AlphaRAD: Grounded Zero-Shot Classification in Chest Radiology via $α$-Corrected Binary Cross Entropy and Factorized Latent Supervision

Vision-Language Pretrained Models (VLPMs) offer a scalable path to open-vocabulary chest radiology understanding, yet two aspects remain underexplored: how structured clinical semantics extracted from medical reports can reduce in-batch noise during contrastive learning, and how cross-modal fusion can be designed to produce more faithful spatial grounding without added complexity. We introduce AlphaRAD, addressing these opportunities through two contributions. First, we construct a large-scale structured medical concept space from medical reports parsed by a Large Language Model for training, thereby mitigating in-batch learning noise and removing heuristic pair matching in contrastive learning, and thus naturally positioning AlphaRAD as a medical concept discriminator trained via $α$-Corrected Binary Cross-Entropy. Second, we propose FLaS (Factorized Latent Supervision), an extremely simple yet effective cross-modal feature fusion module that factorizes VLPM representations into independent subspaces, using dedicated alignment supervision to enhance the expressiveness of spatial grounding without introducing additional model parameters. Through extensive empirical validation, AlphaRAD shows strong zero-shot generalization across diverse chest radiology tasks. Notably, it establishes state-of-the-art average performance across 16 classification benchmarks, while achieving individual state-of-the-art results via distinct gains on 7 grounding/phrase grounding and 3 segmentation datasets.

cs.CV↗

arXiv2026

Highly crystalline superconducting TiN resonators grown on thermally reconstructed sapphire

High quality crystalline growth of a thin film on sapphire requires sufficient substrate preparation, often achieved via the use of aggressive chemical cleaning. Direct thermal reconstruction of the sapphire substrate via a CO$_2$ laser beam may allow for an alternative way to prepare the substrate for epitaxy without the use of any chemical processing. Within this work, we demonstrate that thermal annealing of sapphire into its ($\sqrt{31}$$\times$$\sqrt{31}$)$R$$\pm$9° reconstruction is a valid alternative preparation technique for sapphire substrates. TiN films grown via plasma-assisted molecular beam epitaxy upon these substrates exhibit greater crystallinity than those grown on chemically cleaned sapphire substrates. Superconducting resonators fabricated from these films exhibit similar performance, with many possessing internal quality factors at single photon levels greater than 10$^6$ for both substrate preparation methods.

cond-mat.supr-con↗

arXiv2026

LLaVAFlow: Preserving Latent Alignment Flow for Parameter-Efficient Multimodal Fine-Tuning

While Multimodal Large Language Models (MLLMs) exhibit strong generalization, visual instruction tuning for downstream tasks inevitably causes catastrophic forgetting, impairing overall generalization. While existing methods regulate weight updates to reduce forgetting, they overlook the fundamental cross-modal alignment in MLLMs. Based on prior work and our observations, we argue that cross-modal alignment is implicitly captured in the information-compression trajectory. To preserve the alignment flow embedded in the trajectory, we propose LLaVAFlow, an information-theoretic distillation framework. First, we compress the mutual information between the extracted relations and MLLM embeddings, encouraging a learnable module to produce a refined alignment flow that benefits downstream tasks. Second, we maximize the mutual information between the extracted alignment flows of the pretrained and fine-tuned MLLMs, enabling the transfer of compact alignment information. Extensive experiments show that LLaVAFlow is an effective plug-and-play framework that preserves alignment flow and enhances both downstream performance and generalization.

cs.CV↗

arXiv2026

Trapezoidality of flat arrangement polynomials

Fox's conjecture predicts that the absolute values of the coefficients of the Alexander polynomial of an alternating link form a trapezoidal sequence. Kálmán, Mészáros, and Postnikov gave a new proof of trapezoidality for special alternating links using a polynomial associated with flat vector arrangements. We prove that the flat arrangement polynomial has trapezoidal coefficients for every full-dimensional flat vector arrangement, answering the corresponding open question. Our proof gives a geometric interpretation of this polynomial in terms of convex polytopes. As an application, we prove that the Murasugi--Stoimenow polynomial of every connected Eulerian digraph has trapezoidal coefficients.

math.GT↗

arXiv2026

Grounded Decoding for Autoregressive Speech Enhancement via Adaptive Code-Space Grounding and Local LLM Refinement

Large language model (LLM)-based autoregressive speech enhancement (SE) produces natural speech using learned clean-speech priors, but may hallucinate content unsupported by the input. Deterministic SE better preserves observation-coupled evidence, yet often retains residual noise or local distortion. We propose an evidence-grounded generative SE framework that uses a deterministic estimate as imperfect evidence. A Whisper-guided DPRNN produces an enhanced waveform, which is blended with the observation and tokenized into a discrete evidence sequence. The evidence conditions an autoregressive clean-speech token generator and is reused during decoding through Code-Space Grounding (CSG), which penalizes candidates according to their Hamming distance in the factorized finite-scalar-quantized (FSQ) space. Because the appropriate grounding strength depends on acoustic difficulty, we introduce SNR-Conditioned CSG (SNR-CSG), which maps a calibrated residual-SNR estimate to an utterance-level strength and constructs an adaptive grounded anchor. Although grounding improves content fidelity, the anchor may retain local acoustic defects inherited from the evidence. Since such defects are predominantly local in the FSQ space, nearby tokens may provide better acoustic realizations without large departures from the observation-supported trajectory. We therefore propose Grounded Neighborhood Refinement with LLM ranking (GNR-LLM). It performs one additional teacher-forced pass conditioned on the grounded-anchor history, intersects the LLM top-$K$ candidates with a local FSQ Hamming neighborhood. Experiments on in-domain, controlled-SNR, and DNS no-reverb conditions show that SNR-CSG provides robust automatic grounding, while GNR-LLM substantially improves low-SNR perceptual quality without sacrificing content fidelity.

cs.SD↗

arXiv2026

On the equivalence of $c$-potentiability and $c$-path boundedness in the sense of Artstein-Avidan, Sadovsky, and Wyczesany

A cornerstone of convex analysis, established by Rockafellar in 1966, asserts that a set has a potential if and only if it is cyclically monotone. This characterization was generalized to hold for any {\color{black} finite-valued} cost function $c$ and lies at the core structure of optimal transport plans. However, this equivalence fails to hold for costs that attain infinite values. In this paper, we explore potentiability for an infinite-valued cost $c$ under the assumption of $c$-path boundedness, a condition that was first introduced by Artstein-Avidan, Sadovsky and Wyczesany. This condition is necessary for potentiability and it is more restrictive than $c$-cyclic monotonicity; however, this condition alone does not imply potentiability. We provide general settings and other conditions under which $c$-path boundedness is sufficient for potentiability, and therefore equivalent. We provide a general theorem for potentiability, requiring no topological assumptions on the spaces or the cost. We then provide sufficiency in separable metric spaces and costs that are continuous in their domain. Finally, we introduce the notion of a $c$-path bounded extension and use it to prove the existence of potentials for a special class of costs on $\mathbb{R}^2$. We illustrate our discussion and results with several examples.

math.OC↗

arXiv2026

Roughness-controlled layer in oscillatory turbulent boundary layers over densely packed uniform roughness

In coastal wave boundary layers over gravel-scale roughness, with near-bed orbital excursions ten to a hundred times the roughness height, the boundary layer is only a few roughness heights thick. A roughness-controlled layer (RCL) of the steady-flow extent two to five element heights would then leave no room for a logarithmic layer, yet experiments over densely packed marbles recover logarithmic profiles within millimetres of the crests. We resolve this contradiction by re-analysing previous Particle Image Velocimetry (PIV) records with a triple decomposition that separates the marble-locked dispersive motion from the stochastic turbulence, across eleven wave, current and wave-current conditions. The boundary layer organises into an RCL, a transition region, and a logarithmic profile layer, with the RCL only one to two tenths of a marble diameter deep. This thinness is kinematic: above a periodic bed, the dispersive field decays over a length fixed by the element spacing, so close packing caps the layer at a fraction of a diameter. The layer is destroyed and rebuilt every half-cycle, tracking the near-bed velocity quasi-steadily, while the eddies within it stay locked to the inter-crest gap. Thinness does not imply weakness: within the layer, the dispersive kinetic energy rivals the turbulent kinetic energy, and the dispersive stress matches, near the crests exceeds, the Reynolds stress, showing that separated wakes carry organised momentum. The logarithmic layer survives because the decay length imposed by the packing is far smaller than the boundary-layer thickness, a margin set by the bed geometry rather than by the forcing.

physics.flu-dyn↗

arXiv2026

StartupBench: Benchmarking General-Purpose Agents on Market-Validated End-to-End Workflows

Recent advances in Large Language Models(LLMs) and agents have substantially improved the ability of AI systems to execute complex tasks. Yet existing benchmarks largely rely on researcher-selected tasks, leaving uncertain whether such progress extends to the work that real-world users actually demand from AI systems. We introduce \textbf{StartupBench}, an E2E agent benchmark grounded in market-validated AI startup products. Rather than defining tasks from pre-defined assumptions about useful agent capabilities, we systematically study AI products with demonstrated adoption, together with their product workflows and users, to identify real-world tasks for which AI has established practical demand across diverse professional domains. We translate these workflows into complete deliverable-oriented tasks and evaluate them with fine-grained rubrics capturing their complex requirements. Across representative models evaluated under a unified agent harness, even the strongest model successfully completes only approximately 30\% of StartupBench, despite making substantial partial progress on many tasks. Further analysis identifies aspects like complex instruction following and domain-specific expertise as major sources of failure. Our results reveal that many market-validated workflows remain beyond the reliable capabilities of current general-purpose agents, establishing StartupBench as an empirical measure of progress toward E2E completions of real-world user tasks.

cs.AI↗

arXiv2026

Partial FC: Training 10 Million Identities on a Single Machine

Training face recognition models with millions of identities is challenging because classifier storage, logit memory, and computation grow linearly with the number of classes, eventually making full softmax impractical even when the backbone itself fits comfortably in memory. We present Partial FC (PFC), a scalable approximation to large-class softmax that preserves every positive class center while activating only a sampled subset of negative centers in each mini-batch. This asymmetric treatment retains every target term while avoiding exhaustive interaction with millions of mostly uninformative negatives. Our distributed implementation partitions the classifier across GPUs and samples within each owned shard, so sampling reduces local matrix multiplication and logit storage while sharding eliminates class-gradient synchronization across workers. Together, these properties reduce GPU-resident classifier memory, computation, and class-dependent communication without feature-based hard-negative retrieval. End-to-end system benchmarks demonstrate efficient scaling to massive class spaces, including 64 million classes on a single eight-GPU machine. Across large-scale face-recognition datasets, moderate sampling rates maintain competitive recognition accuracy while substantially improving training efficiency. Our best PFC configurations achieve 97.2\% TAR on IJB-C at FAR $=10^{-4}$ and 94.0\% TAR on ICCV21-MFR at FAR $=10^{-6}$. Beyond clean training data, PFC is robust to inter-class conflicts, label noise, and long-tailed identity distributions: under 40\% label noise, PFC with conflict filtering raises ICCV21-MFR TAR from 43.9\% to 80.2\%, while on long-tailed data PFC improves TAR from 87.4\% to 92.0\%. These results establish positive-preserving negative sampling as an effective foundation for scalable, accurate, and robust identity classification.

cs.CV↗

arXiv2026

Comprehensive language-image pre-training for 3D medical image understanding

In the 3D medical image domain, vision-language pre-training is used to create vision-language encoders (VLEs) that can support radiologists by retrieving patients with similar abnormalities, predicting likelihoods of abnormality, or, with downstream adaptation, generating radiological reports. While the methodology holds promise, three challenges limit the capabilities of current 3D VLEs: data scarcity due to privacy concerns, high computational costs resulting from the volumetric nature of the images, and a domain shift between the long reports used for training and the short prompts used during inference for, e.g., zero-shot classification. As a consequence, natural-image VLE recipes do not directly transfer to 3D medical imaging. In this paper, we overcome these challenges by injecting additional supervision via a report generation objective and combining vision-language with vision-only pre-training, allowing us to leverage both image-only and paired image-text 3D datasets. Further, we propose a novel loss that addresses the domain shift between long reports and short textual prompts. Through these additional objectives, paired with best practices of the 3D medical imaging domain, we develop the Comprehensive Language-Image Pre-training (COLIPRI) encoder family. Our COLIPRI encoders achieve state-of-the-art performance in report generation, semantic segmentation, classification probing, and zero-shot classification. The model weights and inference code are freely available at https://huggingface.co/microsoft/colipri.

cs.CV↗

arXiv2026

DepthArb: Training-Free Depth-Arbitrated Generation for Occlusion-Robust Image Synthesis

Text-to-image models often struggle to synthesize correct occlusion relationships among multiple objects, especially in densely overlapping regions. Many training-free layout-guided methods enforce 2D spatial constraints but do not explicitly resolve depth-dependent attention competition, which can cause concept mixing and implausible occlusion. To address this problem, we propose DepthArb, a training-free framework that formulates occlusion generation as attention arbitration within a unified denoising trajectory. DepthArb employs two core occlusion-control mechanisms: Attention Arbitration Modulation suppresses background-object attention within foreground support according to relative depth, while Spatial Compactness Control limits attention dispersion to preserve object coherence. Because interference varies during generation, Occlusion Conflict Estimation constructs a shared spatial conflict field to adaptively weight both objectives. Through a unified spatial-text attention interface, DepthArb operates on U-Net cross-attention and the image-to-text component of MMDiT joint attention without model retraining. We further introduce OcclBench, a benchmark with continuous relative-depth specifications and occlusion-specific evaluation metrics. Experiments on OcclBench and public benchmarks show that DepthArb improves several layout and occlusion metrics over the evaluated baselines while maintaining competitive text-image alignment.

cs.CV↗

arXiv2026

Optimal control formulation of transition path problems for Markov Jump Processes

We formulate the transition path problem for Markov jump processes as a stochastic optimal control problem on path space. Transitions between metastable sets are induced by an unbounded terminal cost at a stopping time together with controlled modification of jump rates. The running cost takes an entropic form arising naturally from the Girsanov transform for jump processes, allowing both finite- and infinite-horizon formulations to be expressed as optimal changes of measure relative to a reference process. We prove that the optimal path measure admits an explicit representation in terms of the committor function, which solves an elliptic boundary value problem. The singular control induced by the terminal cost is obtained via $Γ$-convergence, yielding a limiting controlled process whose transition rates correspond to a Doob-h transform. The optimally controlled process generates transition paths almost surely while preserving the bridges of the reference process.

math.OC↗

arXiv2026

CFM-Bench: A Unified Multi-Domain, Multi-Task Benchmark for Channel Foundation Models

Channel foundation models (CFMs) are commonly evaluated in model-specific pipelines that differ in data, radio configurations, partitions, adaptation procedures, task definitions, and metrics, preventing reproducible comparison across CFMs and against task-specific networks. We release CFM-Bench, a unified multi-domain, multi-task benchmark comprising 157,900 official single-frame examples from six domains spanning 3GPP statistical simulation, two ray-tracing pipelines, terrestrial and aerial measurements, and synchronized vehicular multimodal simulation. Source-specific interfaces preserve complex channel state information (CSI) and the physical metadata available in each domain while allowing documented model-specific preprocessing. To reduce spatio-temporal leakage, official partitions isolate complete trajectories, measurement sessions, flights, vehicle links, simulation realizations, or buffered spatial regions. CFM-Bench excludes all benchmark splits from foundation-model pretraining, reserves the official training split for downstream fine-tuning, and reports the data used during model development. Six task groups across PHY, RAN, and ISAC applications cover CSI feedback, frequency and temporal channel extrapolation, propagation-state classification, current- and future-beam prediction, and single-frame and temporal localization. Representative experiments on CSI feedback, channel extrapolation, current-beam prediction, and wireless positioning provide reproducible reference results for pretrained channel-prediction models and task-specific neural networks. These results show that relative model performance can vary across data domains, highlighting the importance of using common data partitions, task definitions, and evaluation metrics. CFM-Bench provides a common substrate for evaluating the transferability of channel representations across models, domains, and tasks.

cs.AI↗

arXiv2026

ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow

In the physical world we inhabit, space and time are fundamentally continuous. However, existing machine learning paradigms for world modeling are largely confined to discrete-time prediction, thereby exhibiting significant inefficiency in capturing the dynamics of physical world. We introduce Physical-Time Flow (PT-Flow), a novel approach that learns a continuous latent velocity field operating in physical time. Crucially, the underlying dynamics of sequential data are parameterized by an ordinary differential equation (ODE) embedded in a well-structured representation space. Under this paradigm, the prediction of future can be recast as temporal integration via an ODE solver in the compressed latent space. Building upon PT-Flow, we construct ODEWorld, a continuous-time latent world model that is both efficient and versatile. By extracting time-variant features and enforcing ODE properties on both the dynamical representation space and the latent velocity field, ODEWorld effectively addresses the long-standing representation collapse issue in latent world model literature. This also enables high-quality image reconstruction even after long-horizon prediction. Moreover, its continuous nature allows for arbitrary temporal resolution and even backward prediction, which is impossible for most discrete-time models. Lastly, ODEWorld can provide rich planning-oriented information to facilitate downstream policy learning. Comprehensive experiments demonstrate that ODEWorld successfully reconciles planning-conducive dynamics abstraction with visual realism, excelling in both video generation and robotic control. Project page: https://dstate.github.io/odeworld_website/.

cs.LG↗

arXiv2026

PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives

Video world models simulate future states conditioned on current observations and user actions. Recent systems have demonstrated impressive video consistency and action controllability over long sequences. However, fairly comparing these interactive models remains challenging. In practice, a human player typically evaluates a world model by pursuing long-horizon objectives through interaction. For example, a user may turn around 360 degrees to see whether the environment remains consistent, or walk into the water and inspect whether realistic water ripples are generated. The action sequence required to achieve the same objective may vary substantially between models, making fixed action-conditioned evaluation unsuitable for cross-model comparison. To address this, we employ multi-modal Agent Players to interact with world models toward specified long-horizon objectives. Building on this paradigm, we introduce PlayWorld, a benchmark providing 171 scenarios, each with a specified objective. To evaluate performance thoroughly, we assess models along four core dimensions: geometry consistency, interaction fidelity, out-of-sight evolution, and insight evolution. In addition, we incorporate basic ability metrics for video quality and controllability. Experiments across nine state-of-the-art world models reveal that current models remain unreliable on long-horizon interactive objectives, particularly in maintaining spatial consistency and persistent state evolution. Code and data are available at https://github.com/kxding/PlayWorld.

cs.CV↗

arXiv2026

Boţ-Nguyen Acceleration, Weighted Mean Ergodic Iteration, and the Beta-Binomial Distribution

In 2023, Boţ and Nguyen introduced a new class of accelerated algorithms for finding a fixed point of a nonexpansive operator as the weak limit of a sequence. In this paper, we analyze a particular instance of their algorithm when the nonexpansive operator is assumed to be linear. Surprisingly, the Boţ-Nguyen acceleration then fits naturally into the framework of weighted mean ergodic iterations. This allows us to identify the weak limit as the projection of the starting point onto the fixed point set. Moreover, the weights involved are closely related to the beta-binomial distribution. Finally, when the parameter is equal to 4, then we obtain strong convergence of the iterates.

math.OC↗

arXiv2026

VOLA: Improving Open-World Driving by VLM-Based Semantic Attribute Prediction

Driving in the real world is open-world: a car may encounter a fallen mattress, a deer, or other objects outside its training data. Naming them is not enough. The system must know how to treat each region: can it drive over it, and how severe would a collision be? We therefore shift scene perception from category labels to dense action-relevant attributes, where each pixel is labeled by how it should affect motion rather than by object name. We instantiate this general formulation with two ordered attributes: 7-rank drivability and 5-rank vulnerability. We read Qwen3.5 image-token hidden states directly as a spatial semantic representation. A lightweight boundary-aware decoder then turns this coarse token grid into sharp full-resolution attribute maps. The whole process requires neither autoregressive text generation nor an external mask model such as SAM. We train on dense attribute labels built in CARLA and test transfer to real scenes and to novel obstacles never seen in training. We compare with vision-only segmenters trained on the same attributes and prompted VLM segmenters. Our model matches strong vision-only segmenters on familiar categories and improves transfer to real open-world anomalies, reaching 69.4% mean vulnerability-rank recall versus 57.1% for the best vision-only baseline and 53.9% for the best prompted VLM baseline. These results show that VLM image tokens provide useful semantic cues for transferring driving attributes to objects outside the training vocabulary.

cs.CV↗