arXiv subjects

Yiqing Wang

Publications and source records attributed to Yiqing Wang.

36 records · Page 2Linked to original sources

arXiv2026

Boosting Medical Visual Understanding From Multi-Granular Language Learning

Recent advances in image-text pretraining have significantly enhanced visual understanding by aligning visual and textual representations. Contrastive Language-Image Pretraining (CLIP) has played a pivotal role in multimodal learning. However, its focus on single-label, single-granularity alignment limits its effectiveness in complex domains such as medical imaging, where images often correspond to multiple high-level labels (e.g., disease categories) across different annotation granularities (e.g., diagnostic description, clinical explanation). To address this, we propose Multi-Granular Language Learning (MGLL), a contrastive learning framework designed to improve both multi-label and cross-granularity alignment. MGLL leverages structured multi-label supervision, integrates textual descriptions across granularities, and introduces soft-label supervision with point-wise constraints to enhance alignment. MGLL employs smooth Kullback-Leibler (KL) divergence to ensure cross-granularity consistency while maintaining computational efficiency as a plug-and-play module for vision-language models. Pretrained on our constructed large-scale multi-granular datasets and evaluated across multiple datasets, MGLL outperforms other state-of-the-art methods in downstream tasks. The code is available at https://github.com/HUANGLIZI/MGLL.

cs.CV↗

arXiv2026

Being-H0.5: Scaling Human-Centric Robot Learning for Cross-Embodiment Generalization

We introduce Being-H0.5, a foundational Vision-Language-Action (VLA) model designed for robust cross-embodiment generalization across diverse robotic platforms. While existing VLAs often struggle with morphological heterogeneity and data scarcity, we propose a human-centric learning paradigm that treats human interaction traces as a universal "mother tongue" for physical interaction. To support this, we present UniHand-2.0, the largest embodied pre-training recipe to date, comprising over 35,000 hours of multimodal data across 30 distinct robotic embodiments. Our approach introduces a Unified Action Space that maps heterogeneous robot controls into semantically aligned slots, enabling low-resource robots to bootstrap skills from human data and high-resource platforms. Built upon this human-centric foundation, we design a unified sequential modeling and multi-task pre-training paradigm to bridge human demonstrations and robotic execution. Architecturally, Being-H0.5 utilizes a Mixture-of-Transformers design featuring a novel Mixture-of-Flow (MoF) framework to decouple shared motor primitives from specialized embodiment-specific experts. Finally, to make cross-embodiment policies stable in the real world, we introduce Manifold-Preserving Gating for robustness under sensory shift and Universal Async Chunking to universalize chunked control across embodiments with different latency and control profiles. We empirically demonstrate that Being-H0.5 achieves state-of-the-art results on simulated benchmarks, such as LIBERO (98.9%) and RoboCasa (53.9%), while also exhibiting strong cross-embodiment capabilities on five robotic platforms.

cs.RO↗

arXiv2026

Low-Cost Architecture and Efficient Pattern Synthesis for Polarimetric Phased Array Based on Polarization Coding Reconfigurable Elements

Polarimetric phased arrays (PPAs) enhance radar target detection and anti-jamming capabilities, but their conventional dual transmit/receive (T/R) channel architecture leads to high cost and system complexity. To address these limitations, this paper proposes a polarization-coding reconfigurable phased array (PCRPA) and associated pattern synthesis techniques, which reduce the channel count while preserving key performance. In the PCRPA, each antenna element connects to a single T/R channel and is equipped with a two-level RF switch, enabling real-time control of its polarization state and subarray grouping. By optimizing both the element polarization codes and the excitation weights, the array can synthesize arbitrarily polarized and dual-polarized beams. Simulation results show that the proposed approach achieves suppressed cross-polarization and comparable sidelobe levels compared to conventional PPAs across a wide scan range, with performance improvements being more pronounced in larger arrays. The inherent channel reduction does, however, incur a trade-off in terms of radiated power and directivity. Experimental validation using an $8\times 8$ X-band array antenna confirms the feasibility and effectiveness of the proposed system. The PCRPA architecture and the accompanying synthesis methods offer a cost-effective solution for large-scale PPA systems, maintaining sidelobe and polarization control with significantly reduced hardware complexity.

eess.SY↗

arXiv2025

LOw-cOst yet High-Performant Sparse Matrix-Matrix Multiplication on Arm SME Architectures

Sparse matrix-dense matrix multiplication (SpMM) is a critical kernel in both scientific computing and emerging graph learning workloads. The recent Armv9 architecture introduces Scalable Matrix Extension (SME), enabling tile-based matrix operations with high throughput. However, effectively exploiting both SME and traditional SIMD resources for unstructured sparse workloads remains an open challenge. To address this, we propose LOOPS, a hybrid execution framework that combines row-wise CSR-part with vector-wise BCSR-part layout, enabling cooperative utilization of vector instructions (NEON) and Scalable Matrix Extension (SME) resources. LOOPS supports multi-precision SpMM across FP64, FP32, and FP16 via an adaptive two-level parallelization scheme guided by a lightweight performance model. Experimental results on the entire SuiteSparse on an Apple's M4Pro CPU show that LOOPS achieves average speedups of 9.93$\times$ (FP32)/14.4$\times$ (FP64) against the CPU baseline TACO and 71.3$\times$ (FP32)/54.8$\times$ (FP64) with respect to Armadillo. A comparison of LOOPS running on the same CPU with two GPU methods (cuSPARSE, Magicube) executed on an NVIDIA A100 GPU show average speedups for LOOPS between 19.8$\times$ and 33.5$\times$, depending on the precision. Notably, LOOPS delivers significantly better energy efficiency than the GPU codes on the A100 GPU.

cs.DC↗

arXiv2025

Giant octupole moment in magnetic multilayers

Multipole moments serve as order parameters for characterizing higher-order magnetic effects in momentum space, providing a framework to describe diverse magnetic responses by extending the concept of magnetism. In this letter, we introduce a methodology to quantitatively determine the multipole moment contributions in anomalous Hall effect through angle-dependent anomalous Hall current, with explicit incorporation of discrete crystal symmetries. Our technique uniquely enables the investigation of octupole contribution in non-periodic systems, particularly at interfaces and surfaces. Typically, in (Ag$_{2}$Fe$_{5}$)$_{n}$ multilayers with quantum-well-engineered $k$-point selectivity, we observe an octupole-dominated anomalous Hall effect in conventional ferromagnetic materials, through first-principles calculations. These results fundamentally challenge the existing theoretical understanding of the anomalous Hall effect, showing that even the conventional contribution arises not only from the dipole moment (net magnetization). Furthermore, we establish practical control over the octupole contribution through two distinct approaches: interface engineering and magnetic ordering reconfiguration, opening new possibilities for manipulating higher-order transport effects.

cond-mat.mes-hall↗

arXiv2025

Arbitrarily long strings of consecutive primes in special sets

Let $F(x)$ be a function of the form $ \sum_{i=1}^r d_i x^{ρ_i}$ where $d_1,\ldots,d_r\in\mathbb{R}$, $0 \leq ρ_1 < \ldots < ρ_r,$ $ρ_r \not\in \mathbb{Z},ρ_i \in \mathbb{R}$ for $ 1 \leq i \leq r$ and $d_r\not=0$. We prove that sets of the form $\{ n \in \mathbb{N}: \{ F(n) \} \in U \}$ for any non-empty open set $U \subset [0,1)$ contain arbitrarily long strings of consecutive primes.

math.NT↗

arXiv2025

SIS-Challenge: Event-based Spatio-temporal Instance Segmentation Challenge at the CVPR 2025 Event-based Vision Workshop

We present an overview of the Spatio-temporal Instance Segmentation (SIS) challenge held in conjunction with the CVPR 2025 Event-based Vision Workshop. The task is to predict accurate pixel-level segmentation masks of defined object classes from spatio-temporally aligned event camera and grayscale camera data. We provide an overview of the task, dataset, challenge details and results. Furthermore, we describe the methods used by the top-5 ranking teams in the challenge. More resources and code of the participants' methods are available here: https://github.com/tub-rip/MouseSIS/blob/main/docs/challenge_results.md

cs.CV↗

arXiv2025

SHAP Stability in Credit Risk Management: A Case Study in Credit Card Default Model

The increasing development in the consumer credit card market brings substantial regulatory and risk management challenges. The advanced machine learning models applications bring concerns about model transparency and fairness for both financial institutions and regulatory departments. In this study, we evaluate the consistency of one commonly used Explainable AI (XAI) technology, SHAP, for variable explanation in credit card probability of default models via a case study about credit card default prediction. The study shows the consistency is related to the variable importance level and hence provides practical recommendation for credit risk management

q-fin.RM↗

arXiv2025

Cube: A Roblox View of 3D Intelligence

Foundation models trained on vast amounts of data have demonstrated remarkable reasoning and generation capabilities in the domains of text, images, audio and video. Our goal at Roblox is to build such a foundation model for 3D intelligence, a model that can support developers in producing all aspects of a Roblox experience, from generating 3D objects and scenes to rigging characters for animation to producing programmatic scripts describing object behaviors. We discuss three key design requirements for such a 3D foundation model and then present our first step towards building such a model. We expect that 3D geometric shapes will be a core data type and describe our solution for 3D shape tokenizer. We show how our tokenization scheme can be used in applications for text-to-shape generation, shape-to-text generation and text-to-scene generation. We demonstrate how these applications can collaborate with existing large language models (LLMs) to perform scene analysis and reasoning. We conclude with a discussion outlining our path to building a fully unified foundation model for 3D intelligence.

cs.CV↗

arXiv2025

ContextRefine-CLIP for EPIC-KITCHENS-100 Multi-Instance Retrieval Challenge 2025

This report presents ContextRefine-CLIP (CR-CLIP), an efficient model for visual-textual multi-instance retrieval tasks. The approach is based on the dual-encoder AVION, on which we introduce a cross-modal attention flow module to achieve bidirectional dynamic interaction and refinement between visual and textual features to generate more context-aware joint representations. For soft-label relevance matrices provided in tasks such as EPIC-KITCHENS-100, CR-CLIP can work with Symmetric Multi-Similarity Loss to achieve more accurate semantic alignment and optimization using the refined features. Without using ensemble learning, the CR-CLIP model achieves 66.78mAP and 82.08nDCG on the EPIC-KITCHENS-100 public leaderboard, which significantly outperforms the baseline model and fully validates its effectiveness in cross-modal retrieval. The code will be released open-source on https://github.com/delCayr/ContextRefine-Clip

cs.CV↗

arXiv2024

Spectrum of the Laplacian on the Page metric

We numerically construct the spectrum of the Laplacian on Page's inhomogeneous Einstein metric on $\mathbb{CP}^2 \# \overline{\mathbb{CP}}^2$ by reducing the problem to a (singular) Sturm-Liouville problem in one dimension. We perform a perturbative analysis based upon a closely related, exactly solvable problem that strongly supports our results. We also study the spectrum of the Lichnerowicz Laplacian on symmetric traceless transverse two-tensors. The method relies on both the isometries of the Page metric and pseudospectral methods to numerically solve the resulting ODEs.

math.SP↗

arXiv2024

Stampede Alert Clustering Algorithmic System Based on Tiny-Scale Strengthened DETR

A novel crowd stampede detection and prediction algorithm based on Deformable DETR is proposed to address the challenges of detecting a large number of small targets and target occlusion in crowded airport and train station environments. In terms of model design, the algorithm incorporates a multi-scale feature fusion module to enlarge the receptive field and enhance the detection capability of small targets. Furthermore, the deformable attention mechanism is improved to reduce missed detections and false alarms for critical targets. Additionally, a new algorithm is innovatively introduced for stampede event prediction and visualization. Experimental evaluations on the PKX-LHR dataset demonstrate that the enhanced algorithm achieves a 34% performance in small target detection accuracy while maintaining the original detection speed.

cs.SI↗

arXiv2024

INSPIRIT: Optimizing Heterogeneous Task Scheduling through Adaptive Priority in Task-based Runtime Systems

As modern HPC computing platforms become increasingly heterogeneous, it is challenging for programmers to fully leverage the computation power of massive parallelism offered by such heterogeneity. Consequently, task-based runtime systems have been proposed as an intermediate layer to hide the complex heterogeneity from the application programmers. The core functionality of these systems is to realize efficient task-to-resource mapping in the form of Directed Acyclic Graph (DAG) scheduling. However, existing scheduling schemes face several drawbacks to determine task priorities due to the heavy reliance on domain knowledge or failure to efficiently exploit the interaction of application and hardware characteristics. In this paper, we propose INSPIRIT, an efficient and lightweight scheduling framework with adaptive priority designed for task-based runtime systems. INSPIRIT introduces two novel task attributes \textit{inspiring ability} and \textit{inspiring efficiency} for dictating scheduling, eliminating the need for application domain knowledge. In addition, INSPIRIT jointly considers runtime information such as ready tasks in worker queues to guide task scheduling. This approach exposes more performance opportunities in heterogeneous hardware at runtime while effectively reducing the overhead for adjusting task priorities. Our evaluation results demonstrate that INSPIRIT achieves superior performance compared to cutting edge scheduling schemes on both synthesized and real-world task DAGs.

cs.DC↗

arXiv2023

Embedding dimension gaps in sparse codes

We study the open and closed embedding dimensions of a convex 3-sparse code $\mathcal{FP}$, which records the intersection pattern of lines in the Fano plane. We show that the closed embedding dimension of $\mathcal{FP}$ is three, and the open embedding dimension is between four and six, providing the first example of a 3-sparse code with closed embedding dimension three and differing open and closed embedding dimensions. We also investigate codes whose canonical form is quadratic, i.e. ``degree two" codes. We show that such codes are realizable by axis-parallel boxes, generalizing a recent result of Zhou on inductively pierced codes. We pose several open questions regarding sparse and low-degree codes. In particular, we conjecture that the open embedding dimension of certain 3-sparse codes derived from Steiner triple systems grows to infinity.

math.CO↗

arXiv2023

SwinMM: Masked Multi-view with Swin Transformers for 3D Medical Image Segmentation

Recent advancements in large-scale Vision Transformers have made significant strides in improving pre-trained models for medical image segmentation. However, these methods face a notable challenge in acquiring a substantial amount of pre-training data, particularly within the medical field. To address this limitation, we present Masked Multi-view with Swin Transformers (SwinMM), a novel multi-view pipeline for enabling accurate and data-efficient self-supervised medical image analysis. Our strategy harnesses the potential of multi-view information by incorporating two principal components. In the pre-training phase, we deploy a masked multi-view encoder devised to concurrently train masked multi-view observations through a range of diverse proxy tasks. These tasks span image reconstruction, rotation, contrastive learning, and a novel task that employs a mutual learning paradigm. This new task capitalizes on the consistency between predictions from various perspectives, enabling the extraction of hidden multi-view information from 3D medical data. In the fine-tuning stage, a cross-view decoder is developed to aggregate the multi-view information through a cross-attention block. Compared with the previous state-of-the-art self-supervised learning method Swin UNETR, SwinMM demonstrates a notable advantage on several medical image segmentation tasks. It allows for a smooth integration of multi-view information, significantly boosting both the accuracy and data-efficiency of the model. Code and models are available at https://github.com/UCSC-VLAA/SwinMM/.

cs.CV↗

arXiv2023

Discrepancies of subtrees

We study multicolour, oriented and high-dimensional discrepancies of the set of all subtrees of a tree. As our main result, we show that the $r$-colour discrepancy of the subtrees of any tree is a linear function of the number of leaves $\ell$ of that tree. More concretely, we show that it is bounded by $\lceil(r-1)\ell/r\rceil$ from below and $\lceil(r-1)\ell/2\rceil$ from above, and that these bounds are asymptotically sharp. Motivated by this result, we introduce natural notions of oriented and high-dimensional discrepancies and prove bounds for the corresponding discrepancies of the set of all subtrees of a given tree as functions of its number of leaves.

math.CO↗

arXiv2021

On Designing a Two-stage Auction for Online Advertising

For the scalability of industrial online advertising systems, a two-stage auction architecture is widely used to enable efficient ad allocation on a large set of corpus within a limited response time. The current deployed two-stage ad auction usually retrieves an ad subset by a coarse ad quality metric in a pre-auction stage, and then determines the auction outcome by a refined metric in the subsequent stage. However, this simple and greedy solution suffers from serious performance degradation, as it regards the decision in each stage separately, leading to an improper ad selection metric for the pre-auction stage. In this work, we explicitly investigate the relation between the coarse and refined ad quality metrics, and design a two-stage ad auction by taking the decision interaction between the two stages into account. We decouple the design of the two-stage auction by solving a stochastic subset selection problem in the pre-auction stage and conducting a general second price (GSP) auction in the second stage. We demonstrate that this decouple still preserves the incentive compatibility of the auction mechanism. As the proposed formulation of the pre-auction stage is an NP-hard problem, we propose a scalable approximation solution by defining a new subset selection metric, namely Pre-Auction Score (PAS). Experiment results on both public and industrial dataset demonstrate the significant improvement on social welfare and revenue of the proposed two-stage ad auction, than the intuitive greedy two-stage auction and other baselines.

cs.GT↗

arXiv2021

Neural Auction: End-to-End Learning of Auction Mechanisms for E-Commerce Advertising

In e-commerce advertising, it is crucial to jointly consider various performance metrics, e.g., user experience, advertiser utility, and platform revenue. Traditional auction mechanisms, such as GSP and VCG auctions, can be suboptimal due to their fixed allocation rules to optimize a single performance metric (e.g., revenue or social welfare). Recently, data-driven auctions, learned directly from auction outcomes to optimize multiple performance metrics, have attracted increasing research interests. However, the procedure of auction mechanisms involves various discrete calculation operations, making it challenging to be compatible with continuous optimization pipelines in machine learning. In this paper, we design \underline{D}eep \underline{N}eural \underline{A}uctions (DNAs) to enable end-to-end auction learning by proposing a differentiable model to relax the discrete sorting operation, a key component in auctions. We optimize the performance metrics by developing deep models to efficiently extract contexts from auctions, providing rich features for auction design. We further integrate the game theoretical conditions within the model design, to guarantee the stability of the auctions. DNAs have been successfully deployed in the e-commerce advertising system at Taobao. Experimental evaluation results on both large-scale data set as well as online A/B test demonstrated that DNAs significantly outperformed other mechanisms widely adopted in industry.

cs.GT↗