arXiv subjects

Yu Lei

Publications and source records attributed to Yu Lei.

At least 37 records · Page 2Linked to original sources

arXiv2026

A Unified Framework for Modeling Heterogeneous Financial Data via Dual-Granularity Prompting

Recent industrial credit scoring models remain heavily reliant on manually tuned statistical learning methods. Despite their potential, deep learning architectures have struggled to consistently outperform traditional statistical models in industrial credit scoring, largely due to the complexity of heterogeneous financial data and the challenge of modeling evolving creditworthiness. To bridge this gap, we introduce FinLangNet, a novel framework that reformulates credit scoring as a multi-scale sequential learning problem. FinLangNet processes heterogeneous financial data through a dual-module architecture that combines tabular feature extraction with temporal sequence modeling, generating probability distributions of users' future financial behaviors across multiple time horizons. A key innovation is our dual-prompt mechanism within the sequential module, which introduces learnable prompts operating at both feature-level granularity for capturing fine-grained temporal patterns and user-level granularity for aggregating holistic risk profiles. Notably, real world deployment yielded a 6.3 pp improvement in KS, along with a 9.9\% reduction in bad debt rate.

cs.CE↗

arXiv2026

A Mechanistic Analysis of Sim-and-Real Co-Training in Generative Robot Policies

Co-training, which combines limited in-domain real-world data with abundant surrogate data such as simulation or cross-embodiment robot data, is widely used for training generative robot policies. Despite its empirical success, the mechanisms that determine when and why co-training is effective remain poorly understood. We investigate the mechanism of sim-and-real co-training through theoretical analysis and empirical study, and identify two intrinsic effects governing performance. The first, \textbf{``structured representation alignment"}, reflects a balance between cross-domain representation alignment and domain discernibility, and plays a primary role in downstream performance. The second, the \textbf{``importance reweighting effect"}, arises from domain-dependent modulation of action weighting and operates at a secondary level. We validate these effects with controlled experiments on a toy model and extensive sim-and-sim and sim-and-real robot manipulation experiments. Our analysis offers a unified interpretation of recent co-training techniques and motivates a simple method that consistently improves upon prior approaches. More broadly, our aim is to examine the inner workings of co-training and to facilitate research in this direction.

cs.RO↗

arXiv2026

What-Meets-Where: Unified Learning of Action and Contact Localization in Images

People control their bodies to establish contact with the environment. To comprehensively understand actions across diverse visual contexts, it is essential to simultaneously consider \textbf{what} action is occurring and \textbf{where} it is happening. Current methodologies, however, often inadequately capture this duality, typically failing to jointly model both action semantics and their spatial contextualization within scenes. To bridge this gap, we introduce a novel vision task that simultaneously predicts high-level action semantics and fine-grained body-part contact regions. Our proposed framework, PaIR-Net, comprises three key components: the Contact Prior Aware Module (CPAM) for identifying contact-relevant body parts, the Prior-Guided Concat Segmenter (PGCS) for pixel-wise contact segmentation, and the Interaction Inference Module (IIM) responsible for integrating global interaction relationships. To facilitate this task, we present PaIR (Part-aware Interaction Representation), a comprehensive dataset containing 13,979 images that encompass 654 actions, 80 object categories, and 17 body parts. Experimental evaluation demonstrates that PaIR-Net significantly outperforms baseline approaches, while ablation studies confirm the efficacy of each architectural component. The code and dataset will be released upon publication.

cs.CV↗

arXiv2026

DD-CAM: Minimal Sufficient Explanations for Vision Models Using Delta Debugging

We introduce a gradient-free framework for identifying minimal, sufficient, and decision-preserving explanations in vision models by isolating the smallest subset of representational units whose joint activation preserves predictions. Unlike existing approaches that aggregate all units, often leading to cluttered saliency maps, our approach, DD-CAM, identifies a 1-minimal subset whose joint activation suffices to preserve the prediction (i.e., removing any unit from the subset alters the prediction). To efficiently isolate minimal sufficient subsets, we adapt delta debugging, a systematic reduction strategy from software debugging, and configure its search strategy based on unit interactions in the classifier head: testing individual units for models with non-interacting units and testing unit combinations for models in which unit interactions exist. We then generate minimal, prediction-preserving saliency maps that highlight only the most essential features. Our experimental evaluation demonstrates that our approach can produce more faithful explanations and achieve higher localization accuracy than the state-of-the-art CAM-based approaches.

cs.CV↗

arXiv2026

Personalized Safety Alignment for Text-to-Image Diffusion Models

Text-to-image diffusion models have revolutionized visual content generation, yet their deployment is hindered by a fundamental limitation: safety mechanisms enforce rigid, uniform standards that fail to reflect diverse user preferences shaped by age, culture, or personal beliefs. To address this, we propose Personalized Safety Alignment (PSA), a framework that transitions generative safety from static filtration to user-conditioned adaptation. We introduce Sage, a large-scale dataset capturing diverse safety boundaries across 1,000 simulated user profiles, covering complex risks often missed by traditional datasets. By integrating these profiles via a parameter-efficient cross-attention adapter, PSA dynamically modulates generation to align with individual sensitivities. Extensive experiments demonstrate that PSA achieves a calibrated safety-quality trade-off: under permissive profiles, it relaxes over-cautious constraints to enhance visual fidelity, while under restrictive profiles, it enforces state-of-the-art suppression, significantly outperforming static baselines. Furthermore, PSA exhibits superior instruction adherence compared to prompt-engineering methods, establishing personalization as a vital direction for creating adaptive, user-centered, and responsible generative AI. Our code, data, and models are publicly available at https://github.com/M-E-AGI-Lab/PSAlign.

cs.CV↗

arXiv2026

From Context to EDUs: Faithful and Structured Context Compression via Elementary Discourse Unit Decomposition

Managing extensive context remains a critical bottleneck for Large Language Models (LLMs), particularly in applications like long-document question answering and autonomous agents where lengthy inputs incur high computational costs and introduce noise. Existing compression techniques often disrupt local coherence through discrete token removal or rely on implicit latent encoding that suffers from positional bias and incompatibility with closed-source APIs. To address these limitations, we introduce the EDU-based Context Compressor, a novel explicit compression framework designed to preserve both global structure and fine-grained details. Our approach reformulates context compression as a structure-then-select process. First, our LingoEDU transforms linear text into a structural relation tree of Elementary Discourse Units (EDUs) which are anchored strictly to source indices to eliminate hallucination. Second, a lightweight ranking module selects query-relevant sub-trees for linearization. To rigorously evaluate structural understanding, we release StructBench, a manually annotated dataset of 248 diverse documents. Empirical results demonstrate that our method achieves state-of-the-art structural prediction accuracy and significantly outperforms frontier LLMs while reducing costs. Furthermore, our structure-aware compression substantially enhances performance across downstream tasks ranging from long-context tasks to complex Deep Search scenarios.

cs.CL↗

arXiv2025

A DeepSeek-Powered AI System for Automated Chest Radiograph Interpretation in Clinical Practice

A global shortage of radiologists has been exacerbated by the significant volume of chest X-ray workloads, particularly in primary care. Although multimodal large language models show promise, existing evaluations predominantly rely on automated metrics or retrospective analyses, lacking rigorous prospective clinical validation. Janus-Pro-CXR (1B), a chest X-ray interpretation system based on DeepSeek Janus-Pro model, was developed and rigorously validated through a multicenter prospective trial (NCT07117266). Our system outperforms state-of-the-art X-ray report generation models in automated report generation, surpassing even larger-scale models including ChatGPT 4o (200B parameters), while demonstrating reliable detection of six clinically critical radiographic findings. Retrospective evaluation confirms significantly higher report accuracy than Janus-Pro and ChatGPT 4o. In prospective clinical deployment, AI assistance significantly improved report quality scores, reduced interpretation time by 18.3% (P < 0.001), and was preferred by a majority of experts in 54.3% of cases. Through lightweight architecture and domain-specific optimization, Janus-Pro-CXR improves diagnostic reliability and workflow efficiency, particularly in resource-constrained settings. The model architecture and implementation framework will be open-sourced to facilitate the clinical translation of AI-assisted radiology solutions.

cs.AI↗

arXiv2025

Companion Agents: A Table-Information Mining Paradigm for Text-to-SQL

Large-scale Text-to-SQL benchmarks such as BIRD typically assume complete and accurate database annotations as well as readily available external knowledge, which fails to reflect common industrial settings where annotations are missing, incomplete, or erroneous. This mismatch substantially limits the real-world applicability of state-of-the-art (SOTA) Text-to-SQL systems. To bridge this gap, we explore a database-centric approach that leverages intrinsic, fine-grained information residing in relational databases to construct missing evidence and improve Text-to-SQL accuracy under annotation-scarce conditions. Our key hypothesis is that when a query requires multi-step reasoning over extensive table information, existing methods often struggle to reliably identify and utilize the truly relevant knowledge. We therefore propose to "cache" query-relevant knowledge on the database side in advance, so that it can be selectively activated at inference time. Based on this idea, we introduce Companion Agents (CA), a new Text-to-SQL paradigm that incorporates a group of agents accompanying database schemas to proactively mine and consolidate hidden inter-table relations, value-domain distributions, statistical regularities, and latent semantic cues before query generation. Experiments on BIRD under the fully missing evidence setting show that CA recovers +4.49 / +4.37 / +14.13 execution accuracy points on RSL-SQL / CHESS / DAIL-SQL, respectively, with larger gains on the Challenging subset +9.65 / +7.58 / +16.71. These improvements stem from CA's automatic database-side mining and evidence construction, suggesting a practical path toward industrial-grade Text-to-SQL deployment without reliance on human-curated evidence.

cs.CL↗

arXiv2025

Dark matter in ALFALFA galaxies: Investigating galaxy-halo connection

This paper aims to investigate the galaxy-halo connection using a large sample of individual galaxies with $\mathrm{H\,I}$ integrated spectra. We determine their dark matter content by applying a dynamical method based on $\mathrm{H\,I}$ line widths measured with the curve-of-growth technique, together with inclination corrections inferred from optical images. We build a sample of 2453 gas-rich predominantly late-type galaxies spanning a stellar mass range of $10^{8.7}M_\odot$ to $10^{11.4}M_\odot$ by matching them one-to-one with their counterparts from the ALFALFA survey and the TNG100 simulation, ensuring a direct match of stellar mass and $\mathrm{H\,I}$ radius. We generate mock images and mock $\mathrm{H\,I}$ integrated spectra for TNG100 galaxies, and apply the same dynamical method to both ALFALFA and TNG100 mock galaxies to infer their dark matter masses. Across all stellar mass bins, ALFALFA galaxies exhibit lower median dark matter masses than the mock TNG100 simulation results. In each bin, this offset is driven by a tail of galaxies with comparatively low dark matter content, which becomes more prominent toward higher stellar masses. In the highest mass bin ($M_* > 10^{11} M_\odot$), late-type ALFALFA galaxies show a median dark matter mass that is 23% lower than that of their counterparts in the TNG100 dark-matter-only simulation, with 32% of ALFALFA galaxies having $M_\mathrm{DM}(<R_\mathrm{HI})<10^{11.5} M_\odot$, compared to 17% in the mock TNG100 sample. These results suggest that a larger fraction of massive late-type galaxies reside in relatively less massive dark matter haloes than predicted by the TNG100 simulation.

astro-ph.GA↗

arXiv2025

Generative Large-Scale Pre-trained Models for Automated Ad Bidding Optimization

Modern auto-bidding systems are required to balance overall performance with diverse advertiser goals and real-world constraints, reflecting the dynamic and evolving needs of the industry. Recent advances in conditional generative models, such as transformers and diffusers, have enabled direct trajectory generation tailored to advertiser preferences, offering a promising alternative to traditional Markov Decision Process-based methods. However, these generative methods face significant challenges, such as the distribution shift between offline and online environments, limited exploration of the action space, and the necessity to meet constraints like marginal Cost-per-Mille (CPM) and Return on Investment (ROI). To tackle these challenges, we propose GRAD (Generative Reward-driven Ad-bidding with Mixture-of-Experts), a scalable foundation model for auto-bidding that combines an Action-Mixture-of-Experts module for diverse bidding action exploration with the Value Estimator of Causal Transformer for constraint-aware optimization. Extensive offline and online experiments demonstrate that GRAD significantly enhances platform revenue, highlighting its effectiveness in addressing the evolving and diverse requirements of modern advertisers. Furthermore, GRAD has been implemented in multiple marketing scenarios at Meituan, one of the world's largest online food delivery platforms, leading to a 2.18% increase in Gross Merchandise Value (GMV) and 10.68% increase in ROI.

cs.LG↗

arXiv2025

Do Large Language Models Think Like the Brain? Sentence-Level Evidences from Layer-Wise Embeddings and fMRI

Understanding whether large language models (LLMs) and the human brain converge on similar computational principles remains a fundamental and important question in cognitive neuroscience and AI. Do the brain-like patterns observed in LLMs emerge simply from scaling, or do they reflect deeper alignment with the architecture of human language processing? This study focuses on the sentence-level neural mechanisms of language models, systematically investigating how layer-wise representations in LLMs align with the dynamic neural responses during human sentence comprehension. By comparing hierarchical embeddings from 14 publicly available LLMs with fMRI data collected from participants, who were exposed to a naturalistic narrative story, we constructed sentence-level neural prediction models to identify the model layers most significantly correlated with brain region activations. Results show that improvements in model performance drive the evolution of representational architectures toward brain-like hierarchies, particularly achieving stronger functional and anatomical correspondence at higher semantic abstraction levels. These findings advance our understanding of the computational parallels between LLMs and the human brain, highlighting the potential of LLMs as models for human language processing.

cs.CL↗

arXiv2025

ABLE: Using Adversarial Pairs to Construct Local Models for Explaining Model Predictions

Machine learning models are increasingly used in critical applications but are mostly "black boxes" due to their lack of transparency. Local explanation approaches, such as LIME, address this issue by approximating the behavior of complex models near a test instance using simple, interpretable models. However, these approaches often suffer from instability and poor local fidelity. In this paper, we propose a novel approach called Adversarially Bracketed Local Explanation (ABLE) to address these limitations. Our approach first generates a set of neighborhood points near the test instance, x_test, by adding bounded Gaussian noise. For each neighborhood point D, we apply an adversarial attack to generate an adversarial point A with minimal perturbation that results in a different label than D. A second adversarial attack is then performed on A to generate a point A' that has the same label as D (and thus different than A). The points A and A' form an adversarial pair that brackets the local decision boundary for x_test. We then train a linear model on these adversarial pairs to approximate the local decision boundary. Experimental results on six UCI benchmark datasets across three deep neural network architectures demonstrate that our approach achieves higher stability and fidelity than the state-of-the-art.

cs.LG↗

arXiv2025

Jet reorientation revealed by intermittent jet activity in radio galaxy 0954+556

Intermittent jet activity of AGNs is a common phenomenon, whereas significant jet reorientation during episodic jet activity in relatively young radio galaxies are rarely reported. The quasar 0954+556 at redshift of 0.903 is an intriguing source exhibiting an unusual radio jet structure with significantly different jet directions at kpc and pc scales. At kpc scales, images from the VLA exhibit a bright core, a linear jet extending 24 kpc to the northwest, and a discrete jet component 16 kpc to the northeast. At pc scales, images from the VLBA show a two-component structure with a projected separation of 360 pc in the north-south direction. The peculiar structure of 0954+556 might result from jet reorientation. Here, our aim was to investigate the possible mechanism via multiscale and multifrequency deep radio images. We performed VLA and VLBA observations of 0954+556. Together with some existing data in the NRAO data archive, we made multiple VLA images at 1.4-22 GHz and VLBA images at 1.7-43 GHz for various image analyses of the jet structure. We identified the location of the radio core at pc scales, detected the faint counter-jets at both pc and kpc scales for the first time, and revealed a diffuse emission region connecting pc and kpc scale forward jets. Our spectral index distribution and spectral aging analysis indicate that 0954+556 might undergo at least two episodes of jet activity during the current AGN phase. Moreover, pc scale polarization maps display a well-resolved spine-sheath polarization structure. It seems that the jet direction of 0954+556 changed significantly during intermittent jet activity. This may explain the different jet orientations and spectral ages observed from kpc to pc scales. The research provides a strong case that AGN jet direction might change rapidly on timescales of one million years.

astro-ph.HE↗

arXiv2025

RhinoInsight: Improving Deep Research through Control Mechanisms for Model Behavior and Context

Large language models are evolving from single-turn responders into tool-using agents capable of sustained reasoning and decision-making for deep research. Prevailing systems adopt a linear pipeline of plan to search to write to a report, which suffers from error accumulation and context rot due to the lack of explicit control over both model behavior and context. We introduce RhinoInsight, a deep research framework that adds two control mechanisms to enhance robustness, traceability, and overall quality without parameter updates. First, a Verifiable Checklist module transforms user requirements into traceable and verifiable sub-goals, incorporates human or LLM critics for refinement, and compiles a hierarchical outline to anchor subsequent actions and prevent non-executable planning. Second, an Evidence Audit module structures search content, iteratively updates the outline, and prunes noisy context, while a critic ranks and binds high-quality evidence to drafted content to ensure verifiability and reduce hallucinations. Our experiments demonstrate that RhinoInsight achieves state-of-the-art performance on deep research tasks while remaining competitive on deep search tasks.

cs.CL↗

arXiv2025

Developing an Agent-Based Mathematical Model for Simulating Post-Irradiation Cellular Response: A Crucial Component of a Digital Twin Framework for Personalized Radiation Treatment

In this study, we present the Physical-Bio Translator, an agent-based simulation model designed to simulate cellular responses following irradiation. This simulation framework is based on a novel cell-state transition model that accurately reflects the characteristics of irradiated cells. To validate the Physical-Bio Translator, we performed simulations of cell phase evolution, cell phenotype evolution, and cell survival. The results indicate that the Physical-Bio Translator effectively replicates experimental cell irradiation outcomes, suggesting that digital cell irradiation experiments can be conducted via computer simulation, offering a more sophisticated model for radiation biology. This work lays the foundation for developing a robust and versatile digital twin at multicellular or tissue scales, aiming to comprehensively study and predict patient responses to radiation therapy.

physics.med-ph↗

arXiv2025

Resolving Indirect Calls in Binary Code via Cross-Reference Augmented Graph Neural Networks

Binary code analysis is essential in scenarios where source code is unavailable, with extensive applications across various security domains. However, accurately resolving indirect call targets remains a longstanding challenge in maintaining the integrity of static analysis in binary code. This difficulty arises because the operand of a call instruction (e.g., call rax) remains unknown until runtime, resulting in an incomplete inter-procedural control flow graph (CFG). Previous approaches have struggled with low accuracy and limited scalability. To address these limitations, recent work has increasingly turned to machine learning (ML) to enhance analysis. However, this ML-driven approach faces two significant obstacles: low-quality callsite-callee training pairs and inadequate binary code representation, both of which undermine the accuracy of ML models. In this paper, we introduce CupidCall, a novel approach for resolving indirect calls using graph neural networks. Existing ML models in this area often overlook key elements such as data and code cross-references, which are essential for understanding a program's control flow. In contrast, CupidCall augments CFGs with cross-references, preserving rich semantic information. Additionally, we leverage advanced compiler-level type analysis to generate high-quality callsite-callee training pairs, enhancing model precision and reliability. We further design a graph neural model that leverages augmented CFGs and relational graph convolutions for accurate target prediction. Evaluated against real-world binaries from GitHub and the Arch User Repository on x86_64 architecture, CupidCall achieves an F1 score of 95.2%, outperforming state-of-the-art ML-based approaches. These results highlight CupidCall's effectiveness in building precise inter-procedural CFGs and its potential to advance downstream binary analysis and security applications.

cs.CR↗

arXiv2025

From Masks to Worlds: A Hitchhiker's Guide to World Models

This is not a typical survey of world models; it is a guide for those who want to build worlds. We do not aim to catalog every paper that has ever mentioned a ``world model". Instead, we follow one clear road: from early masked models that unified representation learning across modalities, to unified architectures that share a single paradigm, then to interactive generative models that close the action-perception loop, and finally to memory-augmented systems that sustain consistent worlds over time. We bypass loosely related branches to focus on the core: the generative heart, the interactive loop, and the memory system. We show that this is the most promising path towards true world models.

cs.LG↗

arXiv2025

Outraged AI: Large language models prioritise emotion over cost in fairness enforcement

Emotions guide human decisions, but whether large language models (LLMs) use emotion similarly remains unknown. We tested this using altruistic third-party punishment, where an observer incurs a personal cost to enforce fairness, a hallmark of human morality and often driven by negative emotion. In a large-scale comparison of 4,068 LLM agents with 1,159 adults across 796,100 decisions, LLMs used emotion to guide punishment, sometimes even more strongly than humans did: Unfairness elicited stronger negative emotion that led to more punishment; punishing unfairness produced more positive emotion than accepting; and critically, prompting self-reports of emotion causally increased punishment. However, mechanisms diverged: LLMs prioritized emotion over cost, enforcing norms in an almost all-or-none manner with reduced cost sensitivity, whereas humans balanced fairness and cost. Notably, reasoning models (o3-mini, DeepSeek-R1) were more cost-sensitive and closer to human behavior than foundation models (GPT-3.5, DeepSeek-V3), yet remained heavily emotion-driven. These findings provide the first causal evidence of emotion-guided moral decisions in LLMs and reveal deficits in cost calibration and nuanced fairness judgements, reminiscent of early-stage human responses. We propose that LLMs progress along a trajectory paralleling human development; future models should integrate emotion with context-sensitive reasoning to achieve human-like emotional intelligence.

cs.CL↗