arXiv subjects

Jing Li

Publications and source records attributed to Jing Li.

At least 127 records · Page 7Linked to original sources

arXiv2026

Team-Based Self-Play With Dual Adaptive Weighting for Fine-Tuning LLMs

While recent self-training approaches have reduced reliance on human-labeled data for aligning LLMs, they still face critical limitations: (i) sensitivity to synthetic data quality, leading to instability and bias amplification in iterative training; (ii) ineffective optimization due to a diminishing gap between positive and negative responses over successive training iterations. In this paper, we propose Team-based self-Play with dual Adaptive Weighting (TPAW), a novel self-play algorithm designed to improve alignment in a fully self-supervised setting. TPAW adopts a team-based framework in which the current policy model both collaborates with and competes against historical checkpoints, promoting more stable and efficient optimization. To further enhance learning, we design two adaptive weighting mechanisms: (i) a response reweighting scheme that adjusts the importance of target responses, and (ii) a player weighting strategy that dynamically modulates each team member's contribution during training. Initialized from a SFT model, TPAW iteratively refines alignment without requiring additional human supervision. Experimental results demonstrate that TPAW consistently outperforms existing baselines across various base models and LLM benchmarks. Our code is publicly available at https://github.com/lab-klc/TPAW.

cs.CL↗

arXiv2026

Personalizing LLMs with Binary Feedback: A Preference-Corrected Optimization Framework

Large Language Model (LLM) personalization aims to align model behaviors with individual user preferences. Existing methods often focus on isolated user histories, neglecting the essential role of inter-user differences. We propose C-BPO, a framework that personalizes LLMs via preference-calibrated binary signals. By treating target user data as positive feedback and other users' data as an auxiliary set of implicit negative signals, C-BPO captures distinct inter-user differences. To mitigate the preference overlap issue, where shared task knowledge is erroneously penalized, we derive an objective grounded in Positive-Unlabeled (PU) learning theory. This approach purifies negative signals by subtracting ``positive bias'', ensuring alignment with unique idiosyncrasies without compromising general helpfulness. Empirical experiments across various personalization tasks and backbone LLMs show C-BPO consistently outperforms baselines, demonstrating the efficacy of preference-calibrated binary signals in modeling inter-user differences.

cs.CL↗

arXiv2026

Vocabulary Hijacking in LVLMs: Unveiling Critical Attention Heads by Excluding Inert Tokens to Mitigate Hallucination

Large Vision-Language Models (LVLMs) have achieved remarkable progress in multimodal tasks, yet their reliability is persistently undermined by hallucinations-generating text that contradicts visual input. Recent studies often attribute these errors to inadequate visual attention. In this work, we analyze the attention mechanisms via the logit lens, uncovering a distinct anomaly we term Vocabulary Hijacking. We discover that specific visual tokens, defined as Inert Tokens, disproportionately attract attention. Crucially, when their intermediate hidden states are projected into the vocabulary space, they consistently decode to a fixed set of unrelated words (termed Hijacking Anchors) across layers, revealing a rigid semantic collapse. Leveraging this semantic rigidity, we propose Hijacking Anchor-Based Identification (HABI), a robust strategy to accurately localize these Inert Tokens. To quantify the impact of this phenomenon, we introduce the Non-Hijacked Visual Attention Ratio (NHAR), a novel metric designed to identify attention heads that remain resilient to hijacking and are critical for factual accuracy. Building on these insights, we propose Hijacking-Aware Visual Attention Enhancement (HAVAE), a training-free intervention that selectively strengthens the focus of these identified heads on salient visual content. Extensive experiments across multiple benchmarks demonstrate that HAVAE significantly mitigates hallucinations with no additional computational overhead, while preserving the model's general capabilities. Our code is publicly available at https://github.com/lab-klc/HAVAE.

cs.MM↗

arXiv2026

PPU-Bench:Real World Benchmark for Personalized Partial Unlearning in Vision Language Models

Multimodal Large Language Models (MLLMs) may memorize sensitive cross-modal information during pretraining. However, existing MLLM unlearning benchmarks rely on synthetic knowledge injection or complete subject-level deletion, which fail to capture realistic, personalized deletion requests that require fine-grained factual control. In this paper, we introduce PPU-Bench, a real-world and fine-tuning-free benchmark for personalized partial unlearning in MLLMs. PPU-Bench contains 24K multimodal and unimodal samples derived from pre-existing knowledge of 500 public figures under three progressively challenging settings: Complete, Selective, and Personalized unlearning. The benchmark evaluates whether methods can remove target knowledge while preserving non-target facts, model utility, and cross-modal consistency. Extensive experiments show that Complete Unlearning often suppresses visual identity rather than factual knowledge, while Selective and Personalized Unlearning expose significant forget--retain trade-offs and challenges in intra-subject factual boundaries. Robustness analysis under cross-image and prompt-based attacks reveals distinct vulnerabilities across different unlearning settings. Motivated by these findings, we propose Boundary-Aware Optimization (BAO), which explicitly models intra-subject forget-retain boundaries. Experimental results on two representative methods demonstrate that BAO can effectively enforce intra-subject factual boundaries.

cs.CV↗

arXiv2026

IdGlow: Dynamic Identity Modulation for Multi-Subject Generation

Multi-subject image generation requires seamlessly harmonizing multiple reference identities within a coherent scene. However, existing methods relying on rigid spatial masks or localized attention often struggle with the "stability-plasticity dilemma," particularly failing in tasks that require complex structural deformations, such as identity-preserving age transformation. To address this, we present IdGlow, a mask-free, progressive two-stage framework built upon Flow Matching diffusion models. In the supervised fine-tuning (SFT) stage, we introduce task-adaptive timestep scheduling aligned with diffusion generative dynamics: a linear decay schedule that progressively relaxes constraints for natural group composition, and a temporal gating mechanism that concentrates identity injection within a critical semantic window, successfully preserving adult facial semantics without overriding child-like anatomical structures. To resolve attribute leakage and semantic ambiguity without explicit layout inputs, we further integrate a badcase-driven Vision-Language Model (VLM) for precise, context-aware prompt synthesis. In the second stage, we design a Fine-Grained Group-Level Direct Preference Optimization (DPO) with a weighted margin formulation to simultaneously eliminate multi-subject artifacts, elevate texture harmony, and recalibrate identity fidelity towards real-world distributions. Extensive experiments on two challenging benchmarks -- direct multi-person fusion and age-transformed group generation -- demonstrate that IdGlow fundamentally mitigates the stability-plasticity conflict, achieving a superior Pareto balance between state-of-the-art facial fidelity and commercial-grade aesthetic quality.

cs.CV↗

arXiv2026

Sub-kelvin thermal conductivity of substrates and on-chip routing in quantum integrated systems

The development of large-scale quantum systems increasingly relies on the close integration of heterogeneous components such as qubits, control electronics, and readout circuits, making thermal management at cryogenic temperatures a central challenge in such architectures. In this work, we present an experimental thermal study of two building blocks of such systems: the substrate and the on-chip routing. We first investigate the sub-kelvin thermal conductivity of four substrate materials: high-resistivity silicon, low-resistivity silicon, borosilicate, and sapphire. We report that high-resistivity silicon exhibits the highest thermal conductivity among the substrates studied ($5\cdot10^{-2}$~W/m$\cdot$K at 300~mK), while low-resistivity silicon, borosilicate, and sapphire show lower values ($8\cdot10^{-4}$~W/m$\cdot$K, 2$\cdot10^{-3}$~W/m$\cdot$K, and 2$\cdot10^{-3}$~W/m$\cdot$K at 300~mK, respectively). Ballistic conductance evaluation using a finite-element non-equilibrium Green's function approach further allows us to extract the phonon mean free path in each substrate and gives insights into the involved scattering mechanisms. Additionally, we employ a dedicated test vehicle to evaluate the impact of on-chip routing on the thermal conductance of the system. Our measurements with superconducting Nb routing lines reveal that the routing increases the in-plane thermal conductance of the system, but the substrate remains the dominant heat path. These results highlight the critical role of the substrate choice within quantum systems and underscore the importance of function partitioning through 3D integration approaches for more efficient thermal management in quantum architectures.

cond-mat.mes-hall↗

arXiv2026

Rashba engineering at van der Waals interfaces

Two-dimensional transition metal dichalcogenide (TMD) interfaces offer a versatile platform for studying emergent quantum phenomena and enabling novel device functionalities. When distinct TMD monolayers are stacked vertically or laterally stitched, their interfaces can exhibit unique electronic band alignments, giving rise to long-lived interlayer excitons, charge transfer effects, and moiré superlattices with correlated states. Here, we demonstrate that the interface between a large variety of two different epitaxially grown TMD monolayers controls the intensity and sign of the Rashba spin splitting, which is probed using THz spintronic emission. Optimized TMD heterobilayers, such as HfSe$_2$/PtSe$_2$, show enhanced THz emission that surpass the spin-to-charge conversion efficiency of bulk TMDs, confirming the presence of Rashba states with large spin splitting at the interface. By combining spin- and angle-resolved photoemission spectroscopy with density functional theory, we reveal that the electronic hybridization between the two different TMD monolayers gives rise to extended in-gap states with strong Rashba spin-orbit coupling. The choice of TMD layers enables to engineer the sign and strength of spin-to-charge conversion in van der Waals heterobilayers opening up perspectives to build efficient and tunable THz spintronic emitters.

cond-mat.mes-hall↗

arXiv2026

Adaptive Dual-Teacher Distillation with Subnetwork Rectification for Bridging Semantic Gaps in Black-Box Domain Adaptation

Assuming that neither source data nor source model parameters are accessible, black-box domain adaptation (BBDA) represents a highly practical yet challenging setting, where transferable knowledge is limited to the predictions of a black-box source model. Existing approaches exploit such knowledge via pseudo-label refinement or by leveraging vision-language models (ViLs), but they often fail to reconcile the inherent discrepancy between task-specific knowledge from black-box models and language-aligned semantic priors of ViLs, resulting in suboptimal integration and degraded adaptation performance. To address this challenge, we propose adaptive Dual-Teacher Distillation with Subnetwork Rectification (DDSR), a framework that explicitly reconciles these complementary yet inconsistent knowledge sources. DDSR employs an adaptive prediction fusion strategy to integrate predictions from the black-box source model and a ViL, generating reliable pseudo-labels for the target domain. A subnetwork-based regularization mechanism mitigates overfitting to noisy supervision by enforcing output consistency and gradient divergency. Furthermore, progressively improved target predictions iteratively refine both pseudo-labels and ViL prompts, enhancing semantic alignment. Finally, class-wise prototypes are used to further optimize target predictions via self-training. Extensive experiments on multiple benchmark datasets demonstrate that DDSR consistently outperforms state-of-the-art methods, including those with access to source data or source model parameters.

cs.CV↗

arXiv2026

Down-bending Breaks in Galactic Disks Are an Intrinsic Byproduct of Inside-out Growth

The exponential profile has long been hypothesized as the fundamental morphology of galactic disks. The IllustrisTNG simulations reproduce diverse surface-density profiles: Type I (single exponential), Type II (down-bending), and Type III (up-bending), consistent with observed mass-size relations and kinematics. Type II disks dominate the stellar-mass regime $M_\star < 10^{10.6} M_\odot$ with a prevalence of about 40%, exhibiting systematically extended morphologies. Conversely, Type III and Type I galaxies are more compact while following the same mass-size scaling relation. Evolutionary histories show that Type II galaxies experience minimal external perturbations, suggesting that Type II disks represent an intrinsic disk form and challenging conventional single-exponential paradigms. We demonstrate that Type II breaks arise naturally via inside-out growth since $z=1$, governed by synchronized cold-gas accretion and localized peaks in specific star formation rate. This mechanism also produces the characteristic U-shaped age profiles of Type II disks. Stellar dynamical redistribution plays a minor role in their formation.

astro-ph.GA↗

arXiv2026

When Model Editing Meets Service Evolution: A Knowledge-Update Perspective for Service Recommendation

The rapid evolution of software services poses substantial challenges to the design and implementation of effective recommendation systems. Traditional service recommendation approaches often rely on static representations and historical usage data, which are insufficient for adapting to the dynamic and evolving nature of service ecosystems. Recently, large language models (LLMs) have shown strong potential to overcome these limitations by leveraging rich contextual understanding. However, their practical use faces two major challenges: outdated service facts and invalid or redundant services. To address these issues, we propose EVOREC, an evolution-aware framework for service recommendation that leverages model editing in a locate-then-edit paradigm to incorporate updated service facts without costly retraining efficiently. This allows the model to remain aligned with evolving service ecosystems. To address invalid service issues, we introduce a Finite Automata (FA)-based constrained decoding mechanism with deduplication, which enforces structural and semantic validity while eliminating repeated services. Experiments on real-world service datasets demonstrate that our framework consistently outperforms existing baselines, e.g., achieving an average relative improvement of 25.9% in Recall@5. Moreover, under evolving service scenarios, our approach outperforms model fine-tuning approaches by 22.3%, demonstrating strong adaptability to service evolution and providing a practical solution for service recommendation in dynamic ecosystems

cs.SE↗

arXiv2026

The PHANGS-Hα survey. Ground-based narrow-band imaging of nearby star-forming galaxies

We present PHANGS-Hα, a narrow-band imaging survey that maps Hα emission over a sample of 65 nearby massive star-forming galaxies. The data were obtained using the MPG-ESO 2.2-meter telescope at La Silla and the du Pont 2.5-meter telescope at Las Campanas Observatory, in the framework of the multi-wavelength cloud-scale (50-100 pc) resolution mapping of molecular gas and star formation conducted by the Physics at High Angular resolution in Nearby GalaxieS (PHANGS) collaboration. PHANGS-Hα complements the already published PHANGS-ALMA, PHANGS-MUSE, PHANGS-HST, and PHANGS-JWST surveys, providing an anchor point for the photometric and astrometric calibration of these datasets, as well as samples of H ii regions, and star formation rate maps for the bulk of the PHANGS sample. We present observations, data processing, and calibration of the PHANGS-Hα dataset, as well as the procedures used to derive emission-line fluxes from narrow-band imaging. A subset of galaxies with available spectroscopic Ha mapping from the PHANGS-MUSE survey allows for a detailed comparison with the narrow-band photometry presented here. This informs a series of best practices for the processing of narrow-band Hα imaging that we apply to the full dataset.

astro-ph.GA↗

arXiv2026

Persistent Interfacial Topological Hall Effect Demonstrating Electrical Readout of Topological Spin Structures in Insulators

Conventional topological Hall effects (THE) require conducting magnets, leaving insulating systems largely inaccessible. Here we introduce the interfacial topological Hall effect (ITHE), where the noncoplanar spin textures of insulating magnets are imprinted onto an adjacent heavy metal via the magnetic proximity effect (MPE) and detected electrically. In Pt/h-LuFeO3 bilayers, h-LuFeO3 hosts a topological spin structure robust against high magnetic fields, arising from a 120° triangular spin lattice with small spin canting that yields nontrivial topology but minimal magnetization. This generates a giant Hall response in Pt up to 0.5% of the longitudinal resistivity and a Hall-conductivity/magnetization ratio above 2 V^{-1}, clearly distinguishable from the spin Hall Hanle effect background. Field- and temperature-dependent analysis further reveals that Pt nanoclusters inherit topological textures from h-LuFeO3 via MPE. Unlike the conventional THE narrow peak-and-dip features, ITHE in Pt/h-LuFeO3 persists across a broad magnetic field range up to 14 T, demonstrating the exceptional stability of the underlying topological spin structure. This establishes ITHE as a powerful and sensitive probe for topological magnetism in ultrathin insulating films and paves the way for new spintronic applications.

cond-mat.mtrl-sci↗

arXiv2026

D-QRELO: Training- and Data-Free Delta Compression for Large Language Models via Quantization and Residual Low-Rank Approximation

Supervised Fine-Tuning (SFT) accelerates taskspecific large language models (LLMs) development, but the resulting proliferation of finetuned models incurs substantial memory overhead. Delta compression addresses this by retaining a single pre-trained LLM with multiple compressed delta weights. However, existing methods fail on models fine-tuned with largescale datasets. We find that larger SFT data scale amplifies delta parameter magnitude, singular values, and entropy, exacerbating compression errors. To tackle this, we propose DQRELO (Delta Compression via Quantization and Residual Low-Rank), a novel training- and data-free delta compression method. It combines coarse-grained one-bit quantization to capture the dominant structure of the delta, followed by compensated residual low-rank approximation to recover fine-grained details from the smaller residual error. Experiments on various LLMs spanning dense and MoE architectures across multiple domains under this challenging setting demonstrate that DQRELO outperforms existing methods. Moreover, we establish key design principles for delta compression through extensive empirical analysis, demonstrating how task difficulty, architecture, and layer positioning create predictable patterns that can guide optimal compression strategies in production systems.

cs.LG↗

arXiv2026

Cracking the Code of Juxtaposition: Can AI Models Understand the Humorous Contradictions

Recent advancements in large multimodal language models have demonstrated remarkable proficiency across a wide range of tasks. Yet, these models still struggle with understanding the nuances of human humor through juxtaposition, particularly when it involves nonlinear narratives that underpin many jokes and humor cues. This paper investigates this challenge by focusing on comics with contradictory narratives, where each comic consists of two panels that create a humorous contradiction. We introduce the YesBut benchmark, which comprises tasks of varying difficulty aimed at assessing AI's capabilities in recognizing and interpreting these comics, ranging from literal content comprehension to deep narrative reasoning. Through extensive experimentation and analysis of recent commercial or open-sourced large (vision) language models, we assess their capability to comprehend the complex interplay of the narrative humor inherent in these comics. Our results show that even state-of-the-art models still lag behind human performance on this task. Our findings offer insights into the current limitations and potential improvements for AI in understanding human creative expressions.

cs.CL↗

arXiv2026

When 'YES' Meets 'BUT': Can Large Models Comprehend Contradictory Humor Through Comparative Reasoning?

Understanding humor-particularly when it involves complex, contradictory narratives that require comparative reasoning-remains a significant challenge for large vision-language models (VLMs). This limitation hinders AI's ability to engage in human-like reasoning and cultural expression. In this paper, we investigate this challenge through an in-depth analysis of comics that juxtapose panels to create humor through contradictions. We introduce the YesBut (V2), a novel benchmark with 1,262 comic images from diverse multilingual and multicultural contexts, featuring comprehensive annotations that capture various aspects of narrative understanding. Using this benchmark, we systematically evaluate a wide range of VLMs through four complementary tasks spanning from surface content comprehension to deep narrative reasoning, with particular emphasis on comparative reasoning between contradictory elements. Our extensive experiments reveal that even the most advanced models significantly underperform compared to humans, with common failures in visual perception, key element identification, comparative analysis and hallucinations. We further investigate text-based training strategies and social knowledge augmentation methods to enhance model performance. Our findings not only highlight critical weaknesses in VLMs' understanding of cultural and creative expressions but also provide pathways toward developing context-aware models capable of deeper narrative understanding though comparative reasoning.

cs.CV↗

arXiv2026

ProRank: Prompt Warmup via Reinforcement Learning for Small Language Models Reranking

Reranking is fundamental to information retrieval and retrieval-augmented generation, with recent Large Language Models (LLMs) significantly advancing reranking quality. Most current works rely on large-scale LLMs (>7B parameters), presenting high computational costs. Small Language Models (SLMs) offer a promising alternative because of computational efficiency. However, our preliminary quantitative analysis reveals key limitations of SLMs: their representation space is narrow, leading to reduced expressiveness, and they struggle with understanding task prompts without fine-tuning. To address these issues, we introduce a novel two-stage training approach, ProRank, for SLM-based document reranking. We propose using reinforcement learning to improve the understanding of task prompts. Additionally, we introduce fine-grained score learning to enhance representation expressiveness and further improve document reranking quality. Extensive experiments suggest that ProRank consistently outperforms both the most advanced open-source and proprietary reranking models. Notably, our 0.5B ProRank even surpasses powerful LLM reranking models on the BEIR benchmark, establishing that properly trained SLMs can achieve superior document reranking performance while maintaining computational efficiency.

cs.IR↗

arXiv2026

Dataset-Level Metrics Attenuate Non-Determinism: A Fine-Grained Non-Determinism Evaluation in Diffusion Language Models

Diffusion language models (DLMs) have emerged as a promising paradigm for large language models (LLMs), yet the non-deterministic behavior of DLMs remains poorly understood. The existing non-determinism evaluations for LLMs predominantly rely on dataset-level metrics under fixed inference configurations, providing limited insight into how model behavior varies across runs and evaluation conditions. In this work, we show that dataset-level metrics systematically attenuate non-determinism in diffusion language models by aggregating sample-level prediction quality across different runs. As a result, configurations with similar aggregate performance can exhibit substantially different behaviors on individual inputs, leaving fine-grained instability and distinct error patterns uncharacterized. To address this limitation, we conduct a fine-grained evaluation of non-determinism based on sample-level prediction differences across a range of model-related factors-including guidance scale, diffusion steps, and Monte Carlo sampling-as well as system-related factors such as batch size, hardware, and numerical precision. Our analysis reveals that non-determinism in DLMs is pervasive and structured, with code generation exhibiting markedly higher sensitivity to factor-level choices than question answering. To attribute sources of non-determinism evaluation, we introduce Factor Variance Attribution (FVA), a cross-factor analysis metric that decomposes observed non-determinism into variance attributable to different evaluation factor settings. Our findings highlight the need for fine-grained, factor-aware evaluation to enable reliable non-determinism assessment of diffusion language models.

cs.LG↗

arXiv2026

TARAC: Mitigating Hallucination in LVLMs via Temporal Attention Real-time Accumulative Connection

Large Vision-Language Models have demonstrated remarkable capabilities, yet they suffer from hallucinations that limit practical deployment. While various mitigation strategies exist, they often incur high computational overhead or require extensive retraining. In this paper, we address the issue of visual attention decay during generation, a key factor contributing to hallucinations. We propose Temporal Attention Real-time Accumulative Connection (TARAC), a novel training-free framework that dynamically accumulates and re-injects historical attention to sustain visual grounding. Inspired by cognitive reinforcement mechanisms, TARAC operates as a lightweight, plug-and-play module. Extensive experiments across diverse models (e.g., LLaVA, Qwen2-VL) and benchmarks demonstrate that TARAC significantly outperforms state-of-the-art methods. Remarkably, it achieves these gains with negligible inference overhead ($\sim$4\% TPOT increase), compared to the substantial costs of existing training-free baselines. Specifically, TARAC reduces hallucinated sentences by 25.2\% on CHAIR and improves Perception score by +10.65 on MME, validating its effectiveness and efficiency.

cs.CV↗