Searcharxiv⌕ Search

arXiv subjects

Jing Li

Publications and source records attributed to Jing Li.

At least 109 records · Page 6Linked to original sources
arXiv2026

Mining or Synthesis? Rethinking Exploration Efficiency in Iterative Alignment of Mathematical Reasoning

Iterative Direct Preference Optimization (DPO) has emerged as a widely used paradigm for aligning Large Language Models on reasoning tasks. Existing approaches typically rely on Best-of-N sampling ($N\geq8$) to mine positive trajectories from the distribution tail. In this work, we show that in mathematical reasoning, increasing $N$ yields diminishing returns while increasing verifier-induced false-positive risk and the distribution shift required for policy updates. To address this, we introduce PACE (Proximal Alignment via Corrective Exploration), a generation-based corrective framework that replaces exhaustive mining with low-budget exploration ($2\leq N\leq3$). Rather than searching for increasingly rare positive samples, PACE synthesizes high-fidelity preference pairs from failed explorations through corrective hindsight refinement and verification-guided filtering. Empirically, PACE matches or exceeds the performance of DPO-R1 ($N=16$) while using about $1/5$ of the compute, and remains robust under 20\% label corruption, where high-$N$ baselines exhibit substantially higher noise exploitation.

cs.CL↗
arXiv2026

SmartDirector: Keyframe-Conditioned Cinematic Video Generation with Narrative Pacing Control

The narrative quality of a video fundamentally determines its perceptual value. Although existing video generation methods can produce visually appealing content, they predominantly rely on sparse conditioning signals such as text prompts or first/last frames, which limits precise control over narrative structure and temporal pacing. In this paper, we propose SmartDirector, a framework that enhances the narrative capacity of video generation models through multiple keyframes. SmartDirector supports flexible generation scenarios including single-shot generation, multi-shot narrative synthesis, and video extension. The framework operates in two stages: Director-Gen generates a low-resolution video conditioned on the provided keyframes, and Director-SR refines the output by exploiting high-resolution keyframes as semantic anchors to recover fine-grained details. To enable robust multi-keyframe training, we construct a data pipeline that curates single-shot and multi-shot sequences from movies. Extensive experiments demonstrate that SmartDirector substantially outperforms existing state-of-the-art approaches. We will release the code to facilitate further research.

cs.CV↗
arXiv2026

From Knowing to Doing: A Memory-Controlled Benchmark for LLM Trading Agents on Stock Markets

Evaluating whether large language model (LLM) agents can profit in capital markets is increasingly framed as end-to-end trading: place an agent in a historical market, let it trade, and measure portfolio returns. This setup is vulnerable to two evaluation failures. First, long backtests often overlap with the knowledge cutoffs of frontier LLMs, allowing memorized tickers, dates, prices, and market narratives to substitute for investment reasoning. Second, raw returns are a noisy proxy for stock-selection ability, since positive performance may come from market beta, style exposure, or favorable regimes rather than genuine alpha. We introduce KTD-Fin (Knowing-To-Doing Financial Benchmark), an end-to-end stock-market trading benchmark that addresses both issues. KTD-Fin uses a data-side masking protocol to anonymize key identifiers and calendar information consistently across prompts and tools, separating historical market memory from investment decision-making. It also incorporates a Barra-style performance attribution framework that decomposes portfolio returns into market, style, and stock-selection alpha components. Across ten frontier LLM agents evaluated on the Chinese CSI300 over a 2024--2026 window, masking substantially changes agent rationales, pushing them towards anonymized factor-based reasoning. Attribution analysis further shows that LLM agents' cumulative returns under leakage-controlled evaluation are largely explained by passive market and style exposure, with limited evidence of persistent stock-selection alpha. These findings suggest that financial LLM benchmarks should evaluate not only whether an agent makes money, but also whether the source of returns reflects transferable investment skill. We release KTD-Fin as a reproducible template for leakage-controlled and attribution-aware evaluation of LLM trading agents.

cs.AI↗
arXiv2026

A Closer Look on Memorization in Tabular Diffusion Model: A Data-Centric Perspective

Diffusion models have shown strong performance in generating high-quality tabular data, but they carry privacy risks by reproducing exact training samples. While prior work focuses on dataset-level augmentation to reduce memorization, little is known about which individual samples contribute most. We present the first data-centric study of memorization dynamics in tabular diffusion models. We quantify memorization for each real sample based on how many generated samples are flagged as replicas, using a relative distance ratio. Our empirical analysis reveals a heavy-tailed distribution of memorization counts: a small subset of samples contributes disproportionately to leakage, confirmed via sample-removal experiments. To understand this, we divide real samples into top- and non-top-memorized groups and analyze their training-time behaviors. We track when each sample is first memorized and monitor per-epoch memorization intensity (AUC). Memorized samples are memorized slightly earlier and show stronger signals in early training. Based on these insights, we propose DynamicCut, a two-stage, model-agnostic mitigation method: (a) rank samples by epoch-wise intensity, (b) prune a tunable top fraction, and (c) retrain on the filtered dataset. Across multiple tabular datasets and models, DynamicCut reduces memorization with minimal impact on data diversity and downstream performance. It also complements augmentation-based defenses. Furthermore, DynamicCut enables cross-model transferability: high-ranked samples identified from one model (e.g., a diffusion model) are also effective for reducing memorization when removed from others, such as GANs and VAEs.

cs.LG↗
arXiv2026

Mitigating Provenance-Role Collapse in Long-Term Agents via Typed Memory Representation

Long-term memory is essential for persistent LLM agents, yet prevailing architectures store historical interactions as unstructured, flat text. This unconstrained storage induces provenance-role collapse, a critical failure mode where agents suffer from source-monitoring errors. To resolve this cognitive vulnerability at the architectural level, we propose MemIR, a typed Memory Intermediate Representation that operationalizes source monitoring as a structural constraint. MemIR writes long-term memory into grounded atoms that separate raw evidence, retrieval cues, and truth-bearing claims, with factual authorization restricted to supported claim atoms. It then applies multi-route atomic projection and provenance-scoped utilization to transform heterogeneous retrieval hits into claim-centered candidate bundles and a normalized fact interface for answer generation. Experiments on LoCoMo and BEAM-100K demonstrate that MemIR consistently outperforms existing memory baselines, especially on tasks requiring source tracking, temporal grounding, and aggregation of fragmented evidence.

cs.CL↗
arXiv2026

Language Bias in LVLMs: From In-Depth Analysis to Simple and Effective Mitigation

Large Vision-Language Models (LVLMs) extend large language models with visual understanding, but remain vulnerable to hallucination, where outputs are fluent yet inconsistent with images. Recent studies link this issue to language bias-the tendency of LVLMs to over-rely on text while neglecting visual inputs. Yet most analyses remain empirical without uncovering its underlying cause. In this paper, we provide a systematic study of language bias and identify its root in modality misalignment during training. Our analysis shows that both Visual Instruction Tuning (VIT) and Direct Preference Optimization (DPO) often prioritize textual improvements, which may cause LVLMs to overly lean toward language modeling rather than balanced multimodal understanding. To address this, we propose two simple yet effective methods: Language Bias Regularization (LBR) which mitigates language bias through regularization during instruction tuning, and Language Bias Penalty (LBP), which penalizes language bias in the DPO training process. Extensive experiments across diverse models and benchmarks demonstrate the effectiveness of our approach. LBR consistently improves performance on over ten general benchmarks, while LBP significantly reduces hallucination and improves trustworthiness. Together, these methods not only mitigate language bias but also advance the overall alignment of LVLMs, all without introducing any additional data or auxiliary models. Our code is publicly available at https://github.com/lab-klc/LVLM-Language-Bias.

cs.CL↗
arXiv2026

Skill Weaving: Efficient LLM Improvement via Modular Skillpacks

Large language models increasingly require specialization across diverse domains, yet existing approaches struggle to balance multi-domain capacities with strict memory and inference constraints. In this work, we introduce SkillWeave, a modular improvement framework that enables LLMs to specialize under fixed memory budgets. SkillWeave partitions full capabilities of a general-purpose model into skillpacks -- lightweight, domain-specific delta modules -- that reorganize and refine the model's internal knowledge. For efficient deployment, SkillWeave integrates SkillZip to compress skillpacks into compact and inference-ready format, enabling strong multi-domain performance with low-latency execution. On multi-task and agentic benchmarks, a 9B SkillWeave model outperforms several baselines and even surpasses a 32B monolithic LLM, while achieving up to 4x speedup.

cs.AI↗
arXiv2026

Thickness-Dependent Spintronic Terahertz Emission in MBE-Grown PtTe$_2$: From Semiconductor to Type-II Dirac Semimetal

Spintronic terahertz (THz) emitters have established themselves as among the most practical broadband THz sources available, yet their performance remains fundamentally limited by the spin Hall conductivity of the nonmagnetic conversion layer - a quantity that is fixed once the material is chosen. Here, we demonstrate that in PtTe$_2$, a type-II Dirac semimetal within the transition metal dichalcogenide family, this limitation can be circumvented by exploiting the dramatic thickness-driven electronic phase evolution of the material itself. Using molecular beam epitaxy to grow PtTe$_2$ films with single-monolayer precision from 1 to 20 ML, we show that the spintronic THz emission tracks the underlying electronic phase diagram directly: it is absent in the single-layer semiconducting phase, turns on sharply at the semimetal transition near 2 ML, and reaches a peak amplitude six times that of an equivalent Pt reference at 10 ML, before declining at larger thicknesses due to THz reabsorption in the increasingly metallic film. This non-monotonic behavior is inconsistent with a bulk inverse spin Hall mechanism and instead reflects a multi-channel spin-to-charge conversion process in which spin-momentum-locked topological surface states and a thickness-dependent interfacial Rashba splitting both contribute and strengthen as the type-II Dirac band structure develops. First-principles calculations of the interfacial spin accumulation reproduce the experimental trend quantitatively, confirming this physical picture. These findings introduce thickness engineering of van der Waals semimetals as a new and accessible route to optimizing spintronic THz emitters and spin-orbit torques in magnetic memories (SOT-MRAMs), with direct implications for the broader class of dimensionally tunable topological materials.

cond-mat.mtrl-sci↗
arXiv2026

BrepForge: Factorized B-rep Synthesis via Wireframe Composition and Boundary-Conditioned Surface Instantiation

Boundary representation (B-rep) is the de facto standard for modern CAD, yet learning-based B-rep synthesis remains challenging due to the tight coupling between discrete topology and continuous geometry. We observe a fundamental asymmetry in B-reps: while wireframe composition involves high-entropy structural decisions, the interior surface geometry is largely constrained by its boundary loops. Motivated by this observation, we propose BrepForge, a generative framework that factorizes B-rep synthesis into two stages: wireframe composition and boundary-conditioned surface instantiation. In the first stage, a face-aware autoregressive model serializes the wireframe into structured sequences that explicitly encode hierarchical Vertex-Edge-Face (V-E-F) connectivity, yielding a topologically complete scaffold. In the second stage, precise surface geometries are instantiated by incorporating learning-free geometric priors derived from boundaries, transforming the complex synthesis task into a structured refinement process. This factorized approach ensures both topological integrity and geometric precision, effectively addressing the inherent complexities of B-rep modeling. Extensive experiments demonstrate that BrepForge outperforms existing baselines with superior geometric complexity and topological validity.

cs.GR↗
arXiv2026

Modality-Decoupled Online Recursive Editing

Online model editing for multimodal large language models (MLLMs) requires assimilating a stream of corrections under tight compute and memory budgets. Yet editors developed for text-only LLMs often degrade on MLLMs: visually dominant activations skew the statistics that shape updates, causing cross-modal conflict, while sequential writes become entangled in a shared edit space and amplify long-horizon interference, causing inter-edit interference. To address these, we propose M-ORE, a modality-decoupled online recursive editor for lifelong MLLM adaptation. M-ORE is derived from a unified proximal-projection formulation and admits a closed-form update with a Sherman-Morrison recursion, yielding constant per-edit overhead. It maintains module-wise locality statistics for the text stack and the visual projector to avoid visually dominated update shaping and performs continual updates in a fixed orthogonal low-rank edit subspace via a Sherman-Morrison recursion to mitigate long-horizon interference. Experiments on multiple MLLM backbones and online editing benchmarks show that our M-ORE method consistently improves reliability, generality, and locality over strong baselines, while achieving favorable quality-efficiency scaling. Our code is publicly available at https://github.com/lab-klc/M-ORE.

cs.LG↗
arXiv2026

An AI-driven robotic system for two-dimensional hetero-assemblies

Nanomaterials stacked on-demand, such as rotationally assembled two-dimensional (2D) van der Waals (vdW) layered compounds, provides a versatile platform for quantum simulation and the exploration of exotic electronic phases. Currently, however, such nanoassemblies remain largely confined to inefficiency, manually operated process, limiting their potential for probing emergent physical phenomena. There is a pressing need in the field for high-precision, automated assembling techniques, especially for the scalable fabrication of 2D twistronic heterostructures. Here, we present an intelligent automation system dedicated to the fabrication of van der Waals stacks, following the state-of-the-art protocol for dry transfer of exfoliated 2D materials. The system further employs metadata generated from each automated stacking procedure to perform reinforcement learning, thereby continuously bettering its performances. As a concrete demonstration, we fabricate twisted bilayer graphene (TBLG) -- known for its challenging preparation -- and exhibit its unconventional superconductivity near the magic angle. Our work may pave the way for high-throughput fabrication of low-dimensional nanomaterials including twistronic heterostructures, where integrating data mining and artificial intelligence can accelerate the discovery of novel physical phenomena.

cond-mat.mes-hall↗
arXiv2026

Transitivity Meets Cyclicity: Explicit Preference Decomposition for Dynamic Large Language Model Alignment

Standard RLHF relies on transitive scalar rewards, failing to capture the cyclic nature of human preferences. While some approaches like the General Preference Model (GPM) address this, we identify a theoretical limitation: their implicit formulation entangles hierarchy with cyclicity, failing to guarantee dominant solutions. To address this, we propose the Hybrid Reward-Cyclic (HRC) model, which utilizes game-theoretic decomposition to explicitly disentangle preferences into orthogonal transitive (scalar) and cyclic (vector) components. Complementing this, we introduce Dynamic Self-Play Preference Optimization (DSPPO), which treats alignment as a time-varying game to progressively guide the policy toward the Nash equilibrium. Synthetic data experiments further validate HRC's structural superiority in mixed transitive--cyclic settings, where HRC converges faster and achieves higher accuracy than GPM. Experiments on RewardBench 2 demonstrate that HRC consistently improves over both BT and GPM baselines (e.g., +1.23% on Gemma-2B-it). In particular, its superior performance in the Ties domain empirically validates the model's robustness in handling complex, non-strict preferences. Extensive downstream evaluations on AlpacaEval 2.0, Arena-Hard-v0.1, and MT-Bench confirm the efficacy of our framework. Notably, when using Gemma-2B-it as the base preference model, HRC+DSPPO achieves a peak length-controlled win-rate of 44.75% on AlpacaEval 2.0 and 46.8% on Arena-Hard-v0.1, significantly outperforming SPPO baselines trained with BT or GPM. Our code is publicly available at https://github.com/lab-klc/Hybrid-Reward-Cyclic.

cs.CL↗
arXiv2026

Proactive Memory for Ad-Hoc Recall over Streaming Dialogues

Real-world dialogue usually unfolds as an infinite stream. It thus requires bounded-state memory mechanisms to operate within an infinite horizon. However, existing read-then-think memory is fundamentally misaligned with this setting, as it cannot support ad-hoc memory recall while streams unfold. To explore this challenge, we introduce \textbf{STEM-Bench}, the first benchmark for \textbf{ST}reaming \textbf{E}valuation of \textbf{M}emory. It comprises over 14K QA pairs in dialogue streams that assess perception fidelity, temporal reasoning, and global awareness under infinite-horizon constraints. The preliminary analysis on STEM-Bench indicates a critical textit{fidelity-efficiency dilemma}: retrieval-based methods use fragment context, while full-context models incur unbounded latency. To resolve this, we propose \textbf{ProStream}, a proactive memory framework for streaming dialogues built on a hierarchical structure. It enables ad-hoc memory recall on demand by reasoning over continuous streams with multi-granular distillation. Moreover, it employs Adaptive Spatiotemporal Optimization to dynamically optimize retention based on expected utility. It enables a bounded knowledge state for lower inference latency without sacrificing reasoning fidelity. Experiments show ProStream delivers higher reasoning fidelity than prior baselines while maintaining substantially lower latency than full-context alternatives.

cs.AI↗
arXiv2026

TOPOS: High-Fidelity and Efficient Industry-Grade 3D Head Generation

High-fidelity 3D head generation plays a crucial role in the film, animation and video game industries. In industrial pipelines, studios typically enforce a fixed reference topology across all head assets, as such a clean and uniform topology is a prerequisite for production-level rigging, skinning and animation. In this paper, we present TOPOS, a framework tailored for single image conditioned 3D head generation that jointly recovers geometry and appearance under such an industry-standard topology. In contrast to general 3D generative models which produce triangle meshes with inconsistent topology and numerous vertices, hindering semantic correspondence and asset-level reuse, TOPOS generates head meshes with a fixed, studio-style topology, enabling consistent vertex-level correspondence across all generated heads. To model heads under this unified topology, we proposed a novel variational autoencoder structure, termed TOPOS-VAE. Inspired by multi-model large language models (MLLMs), our TOPOS-VAE leverages the Perceiver Resampler to convert input pointclouds sampled from head meshes of diverse topologies into the target reference topology. Building upon TOPOS-VAE's structured latent space, we train a rectified flow transformer, TOPOS-DiT, to efficiently generate high-fidelity head meshes from a single image. We further present TOPOS-Texture, an end-to-end module that produces relightable UV texture maps from the same portrait image via fine-tuning a multimodal image generative model. The generated textures are spatially aligned with the underlying mesh geometry and faithfully preserve high-frequency appearance details. Extensive experiments demonstrate that TOPOS achieves state-of-the-art performance on 3D head generation, surpassing both classical face reconstruction methods and general 3D object generative models, highlighting its effectiveness for digital human creation.

cs.CV↗
arXiv2026

MoCam: Unified Novel View Synthesis via Structured Denoising Dynamics

Generative novel view synthesis faces a fundamental dilemma: geometric priors provide spatial alignment but become sparse and inaccurate under view changes, while appearance priors offer visual fidelity but lack geometric correspondence. Existing methods either propagate geometric errors throughout generation or suffer from signal conflicts when fusing both statically. We introduce MoCam, which employs structured denoising dynamics to orchestrate a coordinated progression from geometry to appearance within the diffusion process. MoCam first leverages geometric priors in early stages to anchor coarse structures and tolerate their incompleteness, then switches to appearance priors in later stages to actively correct geometric errors and refine details. This design naturally unifies static and dynamic view synthesis by temporally decoupling geometric alignment and appearance refinement within the diffusion process. Experiments demonstrate that MoCam significantly outperforms prior methods, particularly when point clouds contain severe holes or distortions, achieving robust geometry-appearance disentanglement.

cs.CV↗
arXiv2026

Observation of hidden altermagnetism in Cs$_{1-δ}$V$_2$Te$_2$O

Altermagnets are characterized by anisotropic band/spin splittings in momentum space, dictated by their spin-space group symmetries. However, the real-space modulations of altermagnetism are often neglected and have not been explored experimentally. Here we combine neutron diffraction, angle-resolved photoemission spectroscopy (ARPES), spin-resolved ARPES and density functional theory to demonstrate that Cs$_{1-δ}$V$_2$Te$_2$O realizes a spatially modulated form of altermagnetism, i.e., hidden altermagnetism. Such a state in Cs$_{1-δ}$V$_2$Te$_2$O results from its G-type antiferromagnetism and two-dimensional electronic states, allowing for the development of spatially alternating altermagnetic layers, whose local spin polarizations are directly verified by spin-resolved ARPES measurements. Our experimental discovery of hidden altermagnetism broadens the scope of unconventional magnetism and opens routes to exploring emergent phenomena from real-space modulations of altermagnetic order.

cond-mat.mtrl-sci↗
arXiv2026

Mitigating Context-Memory Conflicts in LLMs through Dynamic Cognitive Reconciliation Decoding

Large language models accumulate extensive parametric knowledge through pre-training. However, knowledge conflicts occur when outdated or incorrect parametric knowledge conflicts with external knowledge in the context. Existing methods address knowledge conflicts through contrastive decoding, but in conflict-free scenarios, static approaches disrupt output distribution. Other dynamic decoding methods attempt to measure the degree of conflict but still struggle with complex real-world situations. In this paper, we propose a two-stage decoding method called Dynamic Cognitive Reconciliation Decoding (DCRD), to predict and mitigate context-memory conflicts. DCRD first analyzes the attention map to assess context fidelity and predict potential conflicts. Based on this prediction, the input is directed to one of two decoding paths: (1) greedy decoding, or (2) context fidelity-based dynamic decoding. This design enables DCRD to handle conflicts efficiently while maintaining high accuracy and decoding efficiency in conflict-free cases. Additionally, to simulate scenarios with frequent knowledge updates, we constructed ConflictKG, a knowledge conflict QA benchmark. Experiments on four LLMs across six QA datasets show that DCRD outperforms all baselines, achieving state-of-the-art performance.

cs.CL↗
arXiv2026

Automated Detection and Climatological Analysis of Ripple-Scale Gravity Wave Instabilities Using a Squeeze-and-Excitation Convolutional Neural Network

All-sky OH airglow imaging provides two-dimensional observations of mesospheric gravity wave structure near ~87 km altitude. Ripple-scale instability signatures, characterized by 5-15 km horizontal wavelengths and short lifetimes, are particularly difficult to identify consistently using manual inspection. In this study, we develop a reproducible, automated detection framework based on a squeeze-and-excitation convolutional neural network (SE-CNN) trained on 41 x 41 pixel image patches, to identify ripple-scale structures in 512 x 512 pixel all-sky airglow images acquired at Yucca Ridge Field Station (40.7o N, 104.9o W). The time-differenced images are normalized using a robust median-absolute-deviation (MAD) scaling procedure to mitigate star contamination and background variability. The model is trained and validated on manually annotated ripple and non-ripple patches, then evaluated using independent test subsets. The automated detection is performed using a sliding-window approach with spatial and temporal clustering criteria for event definition. At the patch level, the classifier achieves 92\% F1-score with high precision and recall. At the event level, automated detections recover approximately 90\% of manually identified ripple events while identifying additional low-amplitude occurrences. Validated against previous manual identification study, the automated detection catalog enables objective quantification of ripple occurrence frequency, seasonal modulation, and lifetime distributions. By emphasizing methodological transparency, calibration considerations, and validation metrics, this framework establishes a scalable measurement technique for systematic detection of mesospheric instability signatures in long-term airglow image archives.

physics.ao-ph↗