Searcharxiv⌕ Search

arXiv subjects

Xu Chen

Publications and source records attributed to Xu Chen.

At least 73 records · Page 4Linked to original sources
arXiv2026

OneRank: Unified Transformer-Native Ranking Architecture for Multi-Task Recommendation

Multi-task learning (MTL) is essential in recommender systems to enable complementary learning among diverse user feedback. While modern industrial practices have shifted from DNNs to Transformer-centric architectures to strengthen sequence modeling and scaling capacity, they still decouple feature encoding from multi-task prediction, treating the Transformer as a task-agnostic encoder. This design fundamentally limits the performance and scalability by (1) creating an information bottleneck under heterogeneous task objectives, (2) inducing gradient interference that leads to the seesaw phenomenon, and (3) forcing a dataflow transition in which attention-based, context-adaptive representation learning is converted to static feed-forward task prediction with incompatible information read-write dynamics. We propose OneRank, a Transformer-native multi-task ranking framework that eliminates encoder-predictor separation and introduces task-private channels for forward representation learning and backward optimization, enabling task-specialized learning while reducing inter-task interference. In the forward pass, OneRank learns task-specific representations bottom-up through task-conditioned information selection, candidate-aware contextualization, and controlled cross-task interaction. In the backward pass, cross-task gradient detachment isolates task-private parameter updates from shared knowledge extraction modules, preventing negative transfer. We further replace static task-specific MLP scorers with dynamic matching-based scoring for context-aware personalized ranking. By internalizing multi-task reasoning within the Transformer stack, OneRank establishes a unified and scalable architectural paradigm. Offline and online experiments on large-scale industrial datasets show that OneRank significantly outperforms state-of-the-art baselines while maintaining computational efficiency.

cs.IR↗
arXiv2026

Benchmarking LLMs for Community Governance Simulation with Life-history Narratives

Effective community governance hinges on understanding what specific residents think and need. Recent work has used large language models (LLMs) to simulate human respondents, offering a scalable, reproducible way to study human attitudes and behaviors at low cost. However, these studies typically prompt the model with just a few demographic variables (age, gender, income), simulating only general role types. This is insufficient for community governance, where decisions depend on the views of specific residents. We bridge this gap with an integrated research framework covering dataset, benchmark, algorithm, and system. The dataset comprises approximately 1.2 million characters of first-person narrative collected through two-hour semi-structured interviews with each of 92 residents in an urban community, organized around nine community-governance domains. The benchmark probes 18 mainstream LLMs across four prompting strategies and shows that adding rich life-history profiles meaningfully raises fidelity above the no-profile baseline, but this gain comes with more input tokens per call from the longer prompts they require. The algorithm, curriculum-LoRA, is a parameter-efficient personalization framework that, by closing this fidelity-cost gap, matches the strongest baseline's fidelity at roughly 10x lower per-call cost and Pareto-dominates every configuration tested. The system integrates curriculum-LoRA into a closed-loop policy-evaluation pipeline. Together, these results bring individual-level LLM-based resident simulation within reach of resource-constrained local administrations, enabling community-governance decisions to be systematically pre-evaluated in silico before real-world deployment.

cs.CY↗
arXiv2026

Rosetta Memory: Adaptive Memory for Cross-LLM Agents

Memory is the key component for transforming a stateless LLM into a persistent, evolving agent through experience accumulation, long-horizon planning, and continual self-improvement. Existing memory systems typically take the LLM as the center and design memory operations tailored to a specific backbone. In practice, however, users frequently switch between LLMs, for example using Claude for coding and GPT for writing across tasks, or routing different steps to different backbones within a single task for cost-effective trade-offs. As a result, memory written by one model often needs to be consumed by another. Making upstream memory effectively adapt to and activate downstream LLMs remains a critical yet underexplored problem. To bridge this gap, we shift the perspective from LLM-centric memory design to \emph{memory-centric LLM adaptation}. Specifically, we approach the above upstream-downstream memory adaptation problem from both the write and read sides, and design two profile-conditioned operators that are jointly trained to optimize how memory is stored and presented for better task completion. To ensure the learned operators generalize across a broad set of LLMs, we propose a minimum-gain sampling curriculum that prioritizes the least-served LLMs during training. To better measure the operators' actual contribution rather than the LLM's own capability, we design a performance-gap reward that compares against a naive memory baseline. Experiments on HotpotQA, 2WikiMultihopQA, and MuSiQue demonstrate that our model consistently outperforms baselines and remains robust under unseen-model replacement.

cs.LG↗
arXiv2026

DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees

Large Language Model (LLM)-based agents increasingly rely on memory to learn from experiences over continual interactions. However, storing experiences as independent, flat units leads to substantial redundancy and retrieval conflicts, as similar episodes repeat overlapping content and subtle scene variations cause retrieved memories to offer contradictory guidance. To address this, we introduce residual experience, positing that newly acquired experience is often an incremental variation of existing knowledge. We propose DeltaMem, a framework that organizes experience memory into two independent residual trees, one storing goal-conditioned task experience as reusable skills and another for scene-level environment knowledge. Each tree uses a root node for generalized base experiences and incremental delta nodes for subsequent variations, allowing related experiences to share a common foundation without duplication. For retrieval, a failure-penalized similarity scan locates the best match, reconstructing the full experience via root-to-match chain composition. An autonomous consolidation mechanism distills high-frequency paths into new root nodes, enabling the trees to self-organize from general heuristics to specialized variants. Experiments across diverse interactive environments show that DeltaMem consistently outperforms existing baselines. To facilitate future research, we release the code at https://github.com/import-myself/DeltaMem.

cs.AI↗
arXiv2026

SMAC: Spatial-Modal Joint Modeling and Adaptive Representation Collapse for Multimodal Object Tracking

Multimodal multi-object tracking (MOT) under complex illumination remains challenging due to insufficient joint modeling of spatial and modal features and the limited adaptability of fixed fusion strategies. To address these issues, this paper proposes a spatial-modal convolution fusion and distillation-prompt-based multimodal MOT framework. A spatial-modal fusion backbone is first constructed, where a Basic module performs spatial feature extraction and modal interaction via decoupled 3D convolution, while a Mixed module models nonlinear cross-modal correlations through amplitude-phase decomposition. In addition, a representation collapse network is designed for adaptive multimodal fusion. A Distillation Prompt Guidance (DPG) module generates dynamic modal weights under teacher supervision, and a Global Modal Difference Aggregation (GMDA) module preserves discriminative information during multimodal representation collapse. Extensive experiments on the UniRTL dataset demonstrate the effectiveness of the proposed method. The proposed tracker achieves 63.31 HOTA and 79.21 MOTA on the RNT modality, outperforming several state-of-the-art methods while maintaining favorable inference efficiency. The source code and pretrained models are publicly available at https://github.com/QitaiSun/SMAC.

eess.IV↗
arXiv2026

SCL: Towards Domain Generalization via Single-Temporal Multimodal Contrastive Learning for Remote Sensing Change Detection

In recent years, change detection and anomaly detection models based on CNN and transformer have achieved remarkable success across various datasets based on paired data. However, most such methods exhibit limited crossdataset generalization due to domain-specific designs and typically rely on large amounts of paired labeled data. In this paper, based on visual-language pre-training model, we introduce a Single-temporal multimodal Contrastive Learning (SCL) foundation models for change detection without training on the target dataset. To further improve the model's ability to learn context of textual and visual information, we propose a Dynamic Text-vision Context Optimization (DTCO) module for prompt learning. Meanwhile, to address the data dependency issue of existing methods, we introduce a controllable generation and Single-temporal trAINing strategy (SAIN). This allows us to train the model using a large number of existing single-temporal images without the need for paired label. Extensive experiments on various realworld change detection datasets demonstrate the superior performance and generalization of SCL, outperforming state-of-the-art methods under the evaluated settings. Code is available at https://github.com/Kane-Du/scl-cd.git.

cs.CV↗
arXiv2026

TwinQuant: Learnable Subspace Decomposition for 4-Bit LLM Quantization

4-bit quantization reduces the memory footprint and latency of large language model inference, but its aggressive precision reduction can severely degrade accuracy. Prior methods address this by decomposing each weight matrix into two components (e.g., via singular value decomposition) and quantizing them separately, assigning the bulk of values to a low-precision residual component while handling outliers with a high-precision low-rank component. However, such decompositions are designed to minimize the real-valued energy of the residual, rather than the post-quantization error of the residual and low-rank components. We propose TwinQuant, a 4-bit quantization framework that learns quantization-friendly decomposed subspaces and jointly reshapes both the low-rank and residual components. TwinQuant learns component-specific transformations via a joint optimization over the Stiefel and general linear manifolds, flattening their distributions and reducing dynamic-range imbalance. To enable efficient end-to-end execution, we further design a fused dual-component kernel that pipelines the two-stage low-rank computation on-chip and merges both components with a single epilogue, avoiding intermediate global-memory traffic. Across LLaMA3 and Qwen3 models, TwinQuant preserves near-FP16 accuracy and delivers up to $1.8\times$ end-to-end speedup over an FP16 baseline.

cs.DC↗
arXiv2026

Breaking-induced energy dissipation of surface gravity waves at varying scales and co-flowing wind stresses

Breaking-induced energy dissipation is studied for individual unsteady breaking waves using laboratory measurements of unidirectional surface gravity wave groups across a range of wave scales and wind stresses. A refined framework to estimate breaking-induced dissipation $ΔE_{br}$ is proposed that accounts for background dissipation from non-breaking processes. Using this framework, we show that variations in wave scale primarily influence breaking energetics, such as fractional dissipation $ΔE_{br}/E_0$ and dissipation rate $ε_b$, by modifying the breaking onset threshold. Also, co-flowing wind systematically reduces both $ΔE_{br}/E_0$ and $ε_b$ relative to unforced conditions, as wind-forced waves break earlier with reduced crest forward-leaning. Exploiting the crest-front steepness at incipient breaking $\mathcal{S}_{\text{front}}(t_b)$ to characterise breaking onset and local crest geometry, we formulate a scaling for $ε_b$ based on this local measure. This then yields $ΔE_{br}/E_0 \propto β^{*}\,\mathcal{S}_b\,(τ_b/T_b)$, where $β^{*}$ is crest forward leaning, $\mathcal{S}_b$ local steepness, and $τ_b/T_b$ non-dimensional breaking duration. This scaling highlights the important roles of crest asymmetry and breaking duration in setting the breaking energy dissipation. Finally, we consider the breaking strength parameter $b$ by assessing existing steepness-based scaling laws, and relate $b$ to $\mathcal{S}_{\text{front}}(t_b)$, yielding an approximately linear dependence once the breaking-onset threshold is considered.

physics.flu-dyn↗
arXiv2026

Segment-chirped periodically poled lithium niobate waveguides for broadband supercontinuum generation

Supercontinuum generation is a key technology in nonlinear optics, supporting a wide range of applications in frequency metrology and spectroscopy. Integrated photonics offers a promising route toward compact and efficient supercontinuum sources, yet extending the bandwidth while maintaining high spectral flatness remains a central challenge. Here we demonstrate an integrated broadband supercontinuum source based on segment-chirped periodically poled lithium niobate (SC-PPLN) nanophotonic waveguides. By discretizing the chirped poling profile into independently optimized segments, this approach enables high-fidelity ferroelectric domain inversion with near-ideal duty cycles and establishes broadband quasi-phase matching, overcoming the domain inhomogeneity and efficiency limitations commonly encountered in conventional chirped poling. The engineered phase-matching landscape supports efficient wavelength conversion and simultaneous activation of multiple second- and third-order nonlinear processes. Experimentally, we achieve a spectrally flat supercontinuum spanning three optical octaves, from 320 nm in the ultraviolet to 2600 nm in the mid-infrared. These results establish segment-chirped poling as a practical strategy for broadband wavelength conversion and supercontinuum generation in integrated photonics.

physics.optics↗
arXiv2026

On the limiting geometry of unsteady breaking waves subject to co-flowing wind: spectrally-informed versus locally-measured steepness

Wave steepness is a key geometric variable for describing breaking occurrence and its consequences, including energy dissipation and air entrainment. Using three laboratory campaigns under varying spectral conditions and co-flowing wind forcing, we contrast two types of steepness commonly used for unsteady breaking waves: spectrally-informed wave-group steepness (prognostic), obtained from fixed-point surface-elevation records, and locally-measured crest steepness (diagnostic), obtained from spatial surface profiles extracted using the SDBW-I image-processing method developed herein. For the former, the long-adopted $\mathcal{S}_n$ (linear sum of Fourier-component steepness) increases appreciably within about two dominant wavelengths upstream of breaking because of its sensitivity to evolving high-frequency content. When measured sufficiently far upstream, however, wave-group steepness remains approximately linearly related to the local zero-crossing steepness $\mathcal{S}_b$ across bulk unforced conditions. Notwithstanding this, we argue that the crest-front steepness, $\mathcal{S}_{\mathrm{front}}(t_b)$, which delineates the front-face slope at incipient breaking, is the most physically meaningful metric examined here. It exhibits a consistent breaking-onset lower-bound threshold of $\mathcal{S}_{\mathrm{front}}(t_b)\approx0.2$, while values above this threshold decrease with wind speed as crests become less forward leaning. This may be attributed to wind-modified dispersion, enhanced high-frequency spectral content and aerodynamic sheltering, suggesting that wind--wave and wave--wave interactions act as competing mechanisms in triggering breaking through kinematic and energetic processes beyond what geometry alone can explain. Even so, $\mathcal{S}_{\mathrm{front}}(t_b)$ has strong potential as a controlling variable for future studies of breaking energetics and crest-scale dynamics.

physics.flu-dyn↗
arXiv2026

Divergence Meets Consensus: A Multi-Source Negative Sampling Framework for Sequential Recommendation

Negative sampling is significant for training sequential recommendation models under implicit feedback. The predominant strategy, self-guided hard negative sampling, selects negatives based on the model's current state but suffers from three limitations: (1) the coupling between sampling and model updates triggers a vicious cycle that drives the model into local optima; (2) relying on current model parameters narrows sampling to a small region of the item space, reducing diversity and harming generalization; (3) identifying a hard negative requires scoring the entire candidate pool, causing substantial computational overhead with minimal information gain. To address these challenges, we propose MDCNS (Multi-source Divergence-Consensus for Negative Sampling), a novel "Teacher-Peer-Self" framework inspired by Vygotsky's Zone of Proximal Development (ZPD) theory. The proposed method comprises three components, including multi-source scoring, divergence re-ranking, and consensus distillation. Firstly, multi-source scoring incorporates peer and ensemble teacher models to inject external negative signals and break the self-reinforcement loop. Then, divergence re-ranking exploits prediction discrepancy between self and peer models to enhance sampling diversity. Finally, consensus distillation aligns the self model with the teacher via KL divergence, simultaneously improving computational cost utilization. Extensive experiments on six real-world datasets and five backbone models show that MDCNS consistently outperforms state-of-the-art negative sampling methods, demonstrating strong effectiveness and generalization.

cs.IR↗
arXiv2026

L2P: Unlocking Latent Potential for Pixel Generation

Pixel diffusion models have recently regained attention for visual generation. However, training advanced pixel-space models from scratch demands prohibitive computational and data resources. To address this, we propose the Latent-to-Pixel (L2P) transfer paradigm, an efficient framework that directly harnesses the rich knowledge of pre-trained LDMs to build powerful pixel-space models. Specifically, L2P discards the VAE in favor of large-patch tokenization and freezes the source LDM's intermediate layers, exclusively training shallow layers to learn the latent-to-pixel transformation. By utilizing LDM-generated synthetic images as the sole training corpus, L2P fits an already smooth data manifold, enabling rapid convergence with zero real-data collection. This strategy allows L2P to seamlessly migrate massive latent priors to the pixel space using only 8 GPUs. Furthermore, eliminating the VAE memory bottleneck unlocks native 4K ultra-high resolution generation. Extensive experiments across mainstream LDM architectures show that L2P incurs negligible training overhead, yet performs on par with the source LDM on DPG-Bench and reaches 93% performance on GenEval.

cs.CV↗
arXiv2026

Decentralized Rank Scheduling for Energy-Constrained Multi-Task Federated Fine-Tuning in Edge-Assisted IoV Networks

Large-scale Internet of Vehicles (IoV) deployments increasingly demand the on-device adaptation of foundation models to support diverse, mission-critical perception tasks. While federated fine-tuning offers a promising solution for efficient model specialization, existing approaches often struggle to reconcile the inherent conflict between stringent global energy budgets, heterogeneous task demands, and the high volatility of vehicular network connectivity. In this work, we introduce a hierarchical, adaptive framework that decouples multi-task fine-tuning into two interdependent optimization phases. First, we implement a feedback-loop mechanism at the infrastructure level that dynamically redistributes global energy budgets across concurrent tasks based on real-time convergence dynamics and resource utilization. Second, at the vehicle level, we formulate intra-task rank selection as an energy-constrained online learning problem, solved via a novel primal-dual bandit algorithm, UCB-DUAL, which provides theoretical guarantees on sublinear regret. Our approach effectively internalizes global energy constraints into local decision-making, allowing vehicles to autonomously navigate the complex trade-off between model accuracy, latency, and power consumption. Empirical evaluations using a large-scale IoV simulator, driven by real-world trajectory data, confirm that our proposed method significantly outperforms current federated fine-tuning baselines, offering a robust and scalable solution for resource-constrained vehicular intelligence.

cs.LG↗
arXiv2026

CurEvo: Curriculum-Guided Self-Evolution for Video Understanding

Recent advances in self-evolution video understanding frameworks have demonstrated the potential of autonomous learning without human annotations. However, existing methods often suffer from weakly controlled optimization and uncontrolled difficulty progression, as they lack structured guidance throughout the iterative learning process. To address these limitations, we propose CurEvo, a curriculum-guided self-evolution framework that introduces curriculum learning into self-evolution to achieve more structured and progressive model improvement. CurEvo dynamically regulates task difficulty, refines evaluation criteria, and balances data diversity according to model competence, forming a curriculum-guided feedback loop that aligns learning complexity with model capability. Built upon this principle, we develop a multi-dimensional adaptive QA framework that jointly evolves question generation and answer evaluation across perception, recognition, and understanding dimensions, ensuring coherent and measurable curriculum progression. Through this integration, CurEvo transforms weakly controlled self-evolution into a more structured learning process for autonomous video understanding. Across seven backbones, CurEvo consistently improves both benchmark accuracy and evaluator-based semantic score on four VideoQA benchmarks, validating the effectiveness of curriculum-guided self-evolution for video understanding.

cs.CV↗
arXiv2026

Bragg-Williams order competes with superconductivity

Orderings in charge and spin have been extensively studied to unravel their correlation to emergent superconductivity over the past decades. Bragg-Williams order (BWO), a classical structural order parameter describing site occupancy in alloys, has long been speculated to influence superconducting behavior. Yet, its role still remains ambiguous, largely due to the difficulty of isolating BWO from concomitant charge doping or competing electronic instabilities. Here, we establish In2/3PSe3 as a platform wherein indium vacancies are reversibly configurable between ordered and disordered states via thermal treatment. We show that the disordered phase undergoes a pressure-induced superconducting transition with a Tc of 11 K, significantly higher than the 7 K observed in its ordered counterpart. This constitutes a rare instance in which pure BWO variation drives a substantial shift in Tc. By combining a Ginzburg-Landau phenomenological analysis with a BCS-McMillan microscopic description, we demonstrate that BWO naturally suppresses superconductivity through electron-phonon interactions, a mechanism supported by ultra-low-wavenumber Raman measurements. Our findings support BWO as an independent order parameter that competes directly with superconductivity, extending the concept of competing orders beyond conventional electronic and magnetic degrees of freedom.

cond-mat.supr-con↗
arXiv2026

eTFCE: Exact Threshold-Free Cluster Enhancement via Fast Cluster Retrieval

Threshold-free cluster enhancement (TFCE) is widely used for cluster-based inference in neuroimaging, but existing implementations typically rely on discretized approximations that may introduce numerical variability. We present eTFCE, an efficient framework that provides a numerically exact evaluation of the TFCE integral using an optimized cluster retrieval algorithm. Across multiple datasets, eTFCE and the standard implementation produce highly consistent inference results. Voxel-wise comparisons reveal a systematic asymmetry: the standard method yields smaller p-values for more voxels, while eTFCE concentrates stronger statistical evidence within a smaller subset. These differences are primarily confined to voxels near the inference boundary and have minimal impact on overall inference. This pattern is consistent with discretization effects in standard implementations, where the TFCE integral is approximated using a finite set of threshold levels, introducing subtle biases in statistical evidence accumulation across thresholds. Furthermore, eTFCE improves computational efficiency (71.3% of runtime on average) and enables unified computation of multiple cluster-based statistics within a single permutation framework. Overall, eTFCE provides an exact, efficient, and extensible approach to nonparametric neuroimaging inference.

stat.ME↗
arXiv2026

From Coarse to Fine: Self-Adaptive Hierarchical Planning for LLM Agents

Large language model-based agents have recently emerged as powerful approaches for solving dynamic and multi-step tasks. Most existing agents employ planning mechanisms to guide long-term actions in dynamic environments. However, current planning approaches face a fundamental limitation that they operate at a fixed granularity level. Specifically, they either provide excessive detail for simple tasks or insufficient detail for complex ones, failing to achieve an optimal balance between simplicity and complexity. Drawing inspiration from the principle of \textit{progressive refinement} in cognitive science, we propose \textbf{AdaPlan-H}, a self-adaptive hierarchical planning mechanism that mimics human planning strategies. Our method initiates with a coarse-grained macro plan and progressively refines it based on task complexity. It generates self-adaptive hierarchical plans tailored to the varying difficulty levels of different tasks, which can be optimized by imitation learning and capability enhancement. Experimental results demonstrate that our method significantly improves task execution success rates while mitigating overplanning at the planning level, providing a flexible and efficient solution for multi-step complex decision-making tasks. To contribute to the community, our code and data will be made publicly available at https://github.com/import-myself/AHP.

cs.AI↗
arXiv2026

Fast-then-Fine: A Two-Stage Framework with Multi-Granular Representation for Cross-Modal Retrieval in Remote Sensing

Remote sensing (RS) image-text retrieval plays a critical role in understanding massive RS imagery. However, the dense multi-object distribution and complex backgrounds in RS imagery make it difficult to simultaneously achieve fine-grained cross-modal alignment and efficient retrieval. Existing methods either rely on complex cross-modal interactions that lead to low retrieval efficiency, or depend on large-scale vision-language model pre-training, which requires massive data and computational resources. To address these issues, we propose a fast-then-fine (FTF) two-stage retrieval framework that decomposes retrieval into a text-agnostic recall stage for efficient candidate selection and a text-guided rerank stage for fine-grained alignment. Specifically, in the recall stage, text-agnostic coarse-grained representations are employed for efficient candidate selection; in the rerank stage, a parameter-free balanced text-guided interaction block enhances fine-grained alignment without introducing additional learnable parameters. Furthermore, an inter- and intra-modal loss is designed to jointly optimize cross-modal alignment across multi-granular representations. Extensive experiments on public benchmarks demonstrate that the FTF achieves competitive retrieval accuracy while significantly improving retrieval efficiency compared with existing methods.

cs.CV↗