arXiv subjects

Wei Wen

Publications and source records attributed to Wei Wen.

At least 19 recordsLinked to original sources

arXiv2026

Scaling Representation Diversity: Modulated Attention and Reconstructive Regularization for Visual Grounding

Referring Expression Comprehension (REC) is commonly studied under dataset-specific fine-tuning, resulting in specialist models with limited cross-dataset generalization. In this work, we revisit REC from the perspective of unified open-vocabulary grounding and identify representation degeneration as a key obstacle to scaling a single generalist model. To preserve representation diversity, we propose a holistic data-model co-design framework. Architecturally, we introduce the Modulated Attention-Contrastive Head (mACH) for efficient token-level vision-language alignment and a text-conditioned JEPA auxiliary stream that provides complementary gradient support to preserve alignment-active representations without inference overhead. On the data side, we introduce Objects365-Caption, enriching Objects365 with context-aware referring expressions for large-scale language supervision. We further provide a theoretical analysis showing that complementary gradient subspaces preserve alignment capacity and thereby scale representation diversity. Extensive experiments demonstrate that our single-checkpoint framework achieves highly competitive performance on standard REC benchmarks while exhibiting strong generalization across heterogeneous grounding datasets without benchmark-specific adaptation.

cs.CV↗

arXiv2026

Skills Know Their Neighbors: Cluster-Contrastive Capability Pages for Skill Retrieval

As skill libraries grow, large language model agents must retrieve reusable skills from candidates that often share the same topic and vocabulary but implement different capabilities. Retrieval is limited not only by the scorer but also by the text being scored: a document may describe what a skill does without stating which similar requests should be routed elsewhere. We formalize a skill's capability as its \emph{executable region}, the set of queries it can solve, and view its document as a lossy observation of that region. This view exposes a document-imposed component of retrieval error that cannot be removed by improving the retriever alone. We therefore propose \emph{Capability Pages}, cluster-contrastive skill representations containing a positive trigger $\Tpos$, a negative boundary $\Tneg$, and a discriminative body $B$. An offline compiler compares neighboring skills to write these fields. At inference time, the index uses $\Tpos$ and $B$ for candidate recall, while the router uses $\Tneg$ to reject confusable alternatives. On SRA-Bench, which contains 26{,}262 skills and 5{,}400 questions from six datasets, Capability Pages improve Recall@10 for all five tested retrievers, with a mean gain of $2.94$ points. Adding $\Tneg$ to candidate cards improves end-to-end task success by $3.62$ points on average across four executors and six datasets. A transfer evaluation on Chinese SSL-SkillDiscovery reaches $73.07\%$ MRR@50 using the same encoder across conditions. Capability Pages require no modification to the online models; they improve routing by rewriting the offline skill library.

cs.IR↗

arXiv2026

Breaking the Evaluation Paradox: Evaluating High-Entropy Search with Computationally Irreducible Constraints

Evaluating the exhaustive search capabilities of large language models (LLMs) is plagued by a fundamental paradox: verifying completeness requires complete ground truth, yet high-entropy enumeration tasks make such ground truth impossible for humans to create. This causes benchmarks to systematically penalize models for outperforming their human annotators. Despite rapid progress in web-search and deep research agents -- which now issue hundreds of queries, traverse diverse sites, and synthesize long reports -- evaluation still largely relies on partially annotated answer sets, LLM-based judges, or single-answer questions that avoid genuinely exhaustive search scenarios. We break this paradox by shifting the evaluation paradigm from simulating a messy reality to constructing computationally pure challenges. We introduce VERITAS (Verifiable Traversal Assessment for Search), a framework built on the principle of computationally irreducible constraints. By introducing novel, non-optimizable constraints, we create verifiable, sparse-answer search tasks that are computationally equivalent to exhaustive enumeration. These constraints are easy to verify but impossible for LLMs or search engines to optimize, forcing agents to genuinely traverse the entire search space. VERITAS can automatically generate a virtually infinite number of test cases with perfect ground truth and precise difficulty control, with marginal instance cost dominated by hash computations. This provides not only a robust benchmark for evaluating systematic exploration under uncertainty but also a scalable method for generating training data to improve these crucial, yet underdeveloped, capabilities.

cs.IR↗

arXiv2026

Skill Is Not Document: Query-Conditioned Compatibility for LLM Agent Skill Routing

Large language model agents increasingly rely on reusable skills, making skill retrieval a critical front-end component of agent systems. Skill retrieval, however, is not ordinary document retrieval: a useful top-$K$ result must contain individually relevant skills that also form an executable set for the current query. Existing benchmarks and training pipelines largely supervise pairwise relevance and discard the rejection decisions produced when a language model judges a sampled skill combination to be implausible. We introduce R3-Skill, a Chinese--English benchmark that retains these rejections as query-conditioned compatibility supervision. R3-Skill contains 10,246 deduplicated skills, 41,592 accepted queries, and 32,828 rejected annotations across four language directions; all multi-skill test labels were independently reviewed by multiple experts, and 15,962 parseable rejections are organized into an eight-class taxonomy. We further propose a two-stage system composed of R3-Embedding, a multi-positive bi-encoder for large-pool recall, and R3-Reranker, a cross-encoder trained with graded ListNet supervision. Our analysis shows that this signal is stage-dependent, helping cross-encoder reranking while providing no benefit for the tested bi-encoder objective. On R3-Skill, the complete pipeline achieves $75.39\%$ Hit@1, $81.97\%$ NDCG@10, and $33.27\%$ Set-Compat, a $36.6\%$ relative gain over the strongest reranking baseline. It also obtains $83.87\%$ NDCG@10 on SkillRet, demonstrating transfer beyond R3-Skill.

cs.IR↗

arXiv2026

Small Vision-Language Models are Smart Compressors for Long Video Understanding

Adapting Multimodal Large Language Models (MLLMs) for hour-long videos is bottlenecked by context limits. Dense visual streams saturate token budgets and exacerbate the lost-in-the-middle phenomenon. Existing heuristics, like sparse sampling or uniform pooling, blindly sacrifice fidelity by discarding decisive moments and wasting bandwidth on irrelevant backgrounds. We propose Tempo, an efficient query-aware framework compressing long videos for downstream understanding. Tempo leverages a Small Vision-Language Model (SVLM) as a local temporal compressor, casting token reduction as an early cross-modal distillation process to generate compact, intent-aligned representations in a single forward pass. To enforce strict budgets without breaking causality, we introduce Adaptive Token Allocation (ATA). Exploiting the SVLM's zero-shot relevance prior and semantic front-loading, ATA acts as a training-free $O(1)$ dynamic router. It allocates dense bandwidth to query-critical segments while compressing redundancies into minimal temporal anchors to maintain the global storyline. Extensive experiments show our 6B architecture achieves state-of-the-art performance with aggressive dynamic compression (0.5-16 tokens/frame). On the extreme-long LVBench (4101s), Tempo scores 52.3 under a strict 8K visual budget, outperforming GPT-4o and Gemini 1.5 Pro. Scaling to 2048 frames reaches 53.7. Crucially, Tempo compresses hour-long videos substantially below theoretical limits, proving true long-form video understanding relies on intent-driven efficiency rather than greedily padded context windows.

cs.CV↗

arXiv2026

Rethinking Model Efficiency: Multi-Agent Inference with Large Models

Most vision-language models (VLMs) apply a large language model (LLM) as the decoder, where the response tokens are generated sequentially through autoregression. Therefore, the number of output tokens can be the bottleneck of the end-to-end latency. However, different models may require vastly different numbers of output tokens to achieve comparable performance. In this work, we conduct a comprehensive analysis of the latency across different components of VLMs on simulated data. The experiment shows that a large model with fewer output tokens can be more efficient than a small model with a long output sequence. The empirical study on diverse real-world benchmarks confirms the observation that a large model can achieve better or comparable performance as a small model with significantly fewer output tokens. To leverage the efficiency of large models, we propose a multi-agent inference framework that keeps large models with short responses but transfers the key reasoning tokens from the small model when necessary. The comparison on benchmark tasks demonstrates that by reusing the reasoning tokens from small models, it can help approach the performance of a large model with its own reasoning, which confirms the effectiveness of our proposal.

cs.CV↗

arXiv2026

Efficient Universal Perception Encoder

Running AI models on smart edge devices can unlock versatile user experiences, but presents challenges due to limited compute and the need to handle multiple tasks simultaneously. This requires a vision encoder with small size but powerful and versatile representations. We present our method, Efficient Universal Perception Encoder (EUPE), which offers both inference efficiency and universally good representations for diverse downstream tasks. We achieve this by distilling from multiple domain-expert foundation vision encoders. Unlike previous agglomerative methods that directly scale down from multiple teachers to an efficient encoder, we demonstrate the importance of first scaling up to a large proxy teacher and then scaling down from this single teacher. Experiments show that EUPE achieves on-par or better performance than individual domain experts of the same size on diverse task domains and also outperforms previous agglomerative encoders. We release the full family of EUPE models and the code to foster future research.

cs.CV↗

arXiv2026

dTRPO: Trajectory Reduction in Policy Optimization of Diffusion Large Language Models

Diffusion Large Language Models (dLLMs) introduce a new paradigm for language generation, which in turn presents new challenges for aligning them with human preferences. In this work, we aim to improve the policy optimization for dLLMs by reducing the cost of the trajectory probability calculation, thereby enabling scaled-up offline policy training. We prove that: (i) under reference policy regularization, the probability ratio of the newly unmasked tokens is an unbiased estimate of that of intermediate diffusion states, and (ii) the probability of the full trajectory can be effectively estimated with a single forward pass of a re-masked final state. By integrating these two trajectory reduction strategies into a policy optimization objective, we propose Trajectory Reduction Policy Optimization (dTRPO). We evaluate dTRPO on 7B dLLMs across instruction-following and reasoning benchmarks. Results show that it substantially improves the core performance of state-of-the-art dLLMs, achieving gains of up to 9.6% on STEM tasks, up to 4.3% on coding tasks, and up to 3.0% on instruction-following tasks. Moreover, dTRPO exhibits strong training efficiency due to its offline, single-forward nature, and achieves improved generation efficiency through high-quality outputs.

cs.AI↗

arXiv2026

T2S-Bench & Structure-of-Thought: Benchmarking and Prompting Comprehensive Text-to-Structure Reasoning

Think about how human handles complex reading tasks: marking key points, inferring their relationships, and structuring information to guide understanding and responses. Likewise, can a large language model benefit from text structure to enhance text-processing performance? To explore it, in this work, we first introduce Structure of Thought (SoT), a prompting technique that explicitly guides models to construct intermediate text structures, consistently boosting performance across eight tasks and three model families. Building upon this insight, we present T2S-Bench, the first benchmark designed to evaluate and improve text-to-structure capabilities of models. T2S-Bench includes 1.8K samples across 6 scientific domains and 32 structural types, rigorously constructed to ensure accuracy, fairness, and quality. Evaluation on 45 mainstream models reveals substantial improvement potential: the average accuracy on the multi-hop reasoning task is only 52.1%, and even the most advanced model achieves 58.1% node accuracy in end-to-end extraction. Furthermore, on Qwen2.5-7B-Instruct, SoT alone yields an average +5.7% improvement across eight diverse text-processing tasks, and fine-tuning on T2S-Bench further increases this gain to +8.6%. These results highlight the value of explicit text structuring and the complementary contributions of SoT and T2S-Bench. Dataset and eval code have been released at https://t2s-bench.github.io/T2S-Bench-Page/.

cs.CL↗

arXiv2026

Hessian-Enhanced Alternating Frequency/Time method for Computing Resonance Backbone Curves

Computing resonance and anti-resonance backbone curves in complex nonlinear mechanical systems is of high engineering relevance but remains computationally challenging, especially for large finite-element (FE) models. Existing manifold-based approaches often rely on polynomial parameterizations, limiting their effectiveness for general smooth, non-polynomial nonlinearities. To overcome these limitations, we develop a direct optimization framework that employs a Lagrange multiplier formulation to determine the resonance backbone curve on the response surface constrained by the harmonic balance governing equations. Crucially, solving this formulation efficiently requires second-order sensitivity information. Therefore, the primary innovation of this work is the derivation of a analytical Hessian Tensor for generic $C^2$-continuous nonlinear elements. This is achieved by combining an extended Alternating Frequency/Time (AFT) method for computing second-order derivatives with local-coordinate tensor transformations. By integrating this analytical Hessian into the solver, the proposed framework ensures robust convergence and significantly reduces runtime, making it practical for large-scale models where numerical differentiation is computationally prohibitive. The method is validated on three benchmarks of increasing complexity: a two-degree-of-freedom (2-DOF) system with cubic nonlinearity, a beam with cubic stiffness or hyperbolic tangent (tanh) friction nonlinearities, and an industrial-scale finite element model of a compressor bladed disk (blisk) with a friction ring damper. Results demonstrate that the proposed framework accurately and efficiently computes both resonance and anti-resonance backbone curves, providing a robust frequency-domain tool for structures with non-polynomial nonlinearities.

physics.class-ph↗

arXiv2026

When should I search more: Adaptive Complex Query Optimization with Reinforcement Learning

Query optimization is a crucial component for the efficacy of Retrieval-Augmented Generation (RAG) systems. While reinforcement learning (RL)-based agentic and reasoning methods have recently emerged as a promising direction on query optimization, most existing approaches focus on the expansion and abstraction of a single query. However, complex user queries are prevalent in real-world scenarios, often requiring multiple parallel and sequential search strategies to handle disambiguation and decomposition. Directly applying RL to these complex cases introduces significant hurdles. Determining the optimal number of sub-queries and effectively re-ranking and merging retrieved documents vastly expands the search space and complicates reward design, frequently leading to training instability. To address these challenges, we propose a novel RL framework called Adaptive Complex Query Optimization (ACQO). Our framework is designed to adaptively determine when and how to expand the search process. It features two core components: an Adaptive Query Reformulation (AQR) module that dynamically decides when to decompose a query into multiple sub-queries, and a Rank-Score Fusion (RSF) module that ensures robust result aggregation and provides stable reward signals for the learning agent. To mitigate training instabilities, we adopt a Curriculum Reinforcement Learning (CRL) approach, which stabilizes the training process by progressively introducing more challenging queries through a two-stage strategy. Our comprehensive experiments demonstrate that ACQO achieves state-of-the-art performance on three complex query benchmarks, significantly outperforming established baselines. The framework also showcases improved computational efficiency and broad compatibility with different retrieval architectures, establishing it as a powerful and generalizable solution for next-generation RAG systems.

cs.AI↗

arXiv2026

Precise Computation of Forced Response Backbone Curves of Frictional Structures Using Analytical Hessian Tensor of Contact Elements

Predicting the forced vibration response of nonlinear mechanical systems with friction is critical for engineering applications. Accurately determining the backbone curve of resonance peaks is pivotal for the design of friction devices. However, the prediction of these curves is computationally challenging owing to the nonconservative and nonsmooth nature of friction nonlinearity. Although techniques such as damped nonlinear normal modes (dNNMs) and phase resonance methods have been applied, they often suffer from convergence issues, and their computational accuracy is compromised under certain conditions. This study proposes a novel method for computing the forced response backbone curves of structures with frictional contact interfaces. The method accurately tracks the backbone curve through a parameter continuation scheme, formulated via Lagrange multipliers and accelerated by incorporating a derived analytical Hessian Tensor of contact elements. This approach yields highly accurate numerical results and enables numerical singularities on the curve to be identified and robustly traversed. The proposed method is validated using an Euler-Bernoulli beam finite-element model and a lumped-parameter blade-damper-blade model. The results demonstrate superior accuracy compared to conventional dNNMs and phase resonance methods, particularly in cases involving either high structural damping or strong frictional damping. This work provides a robust computational tool and presents a detailed comparative analysis that clarifies the applicability and limitations of the proposed and conventional methods.

physics.class-ph↗

arXiv2026

VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice

Chain-of-thought (CoT) reasoning has emerged as a powerful tool for multimodal large language models on video understanding tasks. However, its necessity and advantages over direct answering remain underexplored. In this paper, we first demonstrate that for RL-trained video models, direct answering often matches or even surpasses CoT performance, despite CoT producing step-by-step analyses at a higher computational cost. Motivated by this, we propose VideoAuto-R1, a video understanding framework that adopts a reason-when-necessary strategy. During training, our approach follows a Thinking Once, Answering Twice paradigm: the model first generates an initial answer, then performs reasoning, and finally outputs a reviewed answer. Both answers are supervised via verifiable rewards. During inference, the model uses the confidence score of the initial answer to determine whether to proceed with reasoning. Across video QA and grounding benchmarks, VideoAuto-R1 achieves state-of-the-art accuracy with significantly improved efficiency, reducing the average response length by ~3.3x, e.g., from 149 to just 44 tokens. Moreover, we observe a low rate of thinking-mode activation on perception-oriented tasks, but a higher rate on reasoning-intensive tasks. This suggests that explicit language-based reasoning is generally beneficial but not always necessary.

cs.CV↗

arXiv2025

Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models

We introduce Youtu-LLM, a lightweight yet powerful language model that harmonizes high computational efficiency with native agentic intelligence. Unlike typical small models that rely on distillation, Youtu-LLM (1.96B) is pre-trained from scratch to systematically cultivate reasoning and planning capabilities. The key technical advancements are as follows: (1) Compact Architecture with Long-Context Support: Built on a dense Multi-Latent Attention (MLA) architecture with a novel STEM-oriented vocabulary, Youtu-LLM supports a 128k context window. This design enables robust long-context reasoning and state tracking within a minimal memory footprint, making it ideal for long-horizon agent and reasoning tasks. (2) Principled "Commonsense-STEM-Agent" Curriculum: We curated a massive corpus of approximately 11T tokens and implemented a multi-stage training strategy. By progressively shifting the pre-training data distribution from general commonsense to complex STEM and agentic tasks, we ensure the model acquires deep cognitive abilities rather than superficial alignment. (3) Scalable Agentic Mid-training: Specifically for the agentic mid-training, we employ diverse data construction schemes to synthesize rich and varied trajectories across math, coding, and tool-use domains. This high-quality data enables the model to internalize planning and reflection behaviors effectively. Extensive evaluations show that Youtu-LLM sets a new state-of-the-art for sub-2B LLMs. On general benchmarks, it achieves competitive performance against larger models, while on agent-specific tasks, it significantly surpasses existing SOTA baselines, demonstrating that lightweight models can possess strong intrinsic agentic capabilities.

cs.CL↗

arXiv2025

MobileLLM-Pro Technical Report

Efficient on-device language models around 1 billion parameters are essential for powering low-latency AI applications on mobile and wearable devices. However, achieving strong performance in this model class, while supporting long context windows and practical deployment remains a significant challenge. We introduce MobileLLM-Pro, a 1-billion-parameter language model optimized for on-device deployment. MobileLLM-Pro achieves state-of-the-art results across 11 standard benchmarks, significantly outperforming both Gemma 3-1B and Llama 3.2-1B, while supporting context windows of up to 128,000 tokens and showing only minor performance regressions at 4-bit quantization. These improvements are enabled by four core innovations: (1) implicit positional distillation, a novel technique that effectively instills long-context capabilities through knowledge distillation; (2) a specialist model merging framework that fuses multiple domain experts into a compact model without parameter growth; (3) simulation-driven data mixing using utility estimation; and (4) 4-bit quantization-aware training with self-distillation. We release our model weights and code to support future research in efficient on-device language models.

cs.LG↗

arXiv2025

MobileLLM-R1: Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes

The paradigm shift in large language models (LLMs) from instinctive responses to chain-of-thought (CoT) reasoning has fueled two prevailing assumptions: (1) reasoning capabilities only emerge in sufficiently large models, and (2) such capabilities require training on massive datasets. While the first assumption has already been challenged by recent sub-billion-parameter reasoning models such as Qwen3-0.6B and DeepSeek distilled variants, the second remains largely unquestioned. In this work, we revisit the necessity of scaling to extremely large corpora (>10T tokens) for reasoning emergence. By carefully curating and resampling open-source datasets that we identify as beneficial under our designed metrics, we demonstrate that strong reasoning abilities can emerge with far less data. Specifically, we show that only ~2T tokens of high-quality data are sufficient, and pre-training with 4.2T tokens on the dataset resampled from these ~2T tokens, followed by a established post-training procedure, enables the development of MobileLLM-R1, a series of sub-billion-parameter reasoning models that substantially outperform prior models trained on fully open-sourced data. For example, MobileLLM-R1-950M achieves an AIME score of 15.5, compared to just 0.6 for OLMo-2-1.48B and 0.3 for SmolLM-2-1.7B. Remarkably, despite being trained on only 11.7% of the tokens compared to Qwen3's proprietary 36T-token corpus for pretraining, MobileLLM-R1-950M matches or surpasses Qwen3-0.6B across multiple reasoning benchmarks. To facilitate further research in this direction, we have made the models (https://huggingface.co/collections/facebook/mobilellm-r1) and code (https://github.com/facebookresearch/MobileLLM-R1) publicly available, along with the complete training recipe, data sources, and data mixing ratios.

cs.CL↗

arXiv2025

Relativistic Path-Integral Origin of the Dirac Equation, Quantum Collapse, Decoherence and Non-Hermitian Phenomena

Relativity and quantum mechanics are two cornerstones of modern physics, yet their unification within a single-particle path integral and a dynamical explanation of quantum measurement remain unresolved. Historically, these two problems have been treated as separate, but here we show they are intimately linked. We construct a self-consistent relativistic path integral that yields the Dirac and other standard wave equations under differetialable potentials. More importantly, we find that this propagator contains a latent, nonlocal correlation that is activated by realistic electromagnetic noise. This correlation unifies unitary evolution and wave-function collapse into a single dynamical mechanism: while differentiable potentials preserve unitary driving, nondifferentiable noise activates a bounded-martingale stochastic process that induces collapse. We show that the characteristics of quantum measurement are naturally derived from this stochastic dynamical process, thereby turning the axioms of quantum measurement from postulates into dynamical consequences. Furthermore, averaging this stochastic evolution over the noise record recovers the Gorini-Kossakowski-Sudarshan-Lindblad (GKSL) master equation, providing a first-principles derivation of decoherence free from the method of Born-Markov approximation. Extending this approach to composite systems establishes a stochastic foundation for effective non-Hermitian descriptions while preserving relativistic causality. Finally, because the noise spectrum governs the collapse process, engineering ``colored'' noise can actively accelerate or steer state reduction, suggesting new routes toward fast qubit reset and enhanced quantum control.

physics.gen-ph↗

arXiv2025

ASPD: Unlocking Adaptive Serial-Parallel Decoding by Exploring Intrinsic Parallelism in LLMs

The increasing scale and complexity of large language models (LLMs) pose significant inference latency challenges, primarily due to their autoregressive decoding paradigm characterized by the sequential nature of next-token prediction. By re-examining the outputs of autoregressive models, we observed that some segments exhibit parallelizable structures, which we term intrinsic parallelism. Decoding each parallelizable branch simultaneously (i.e. parallel decoding) can significantly improve the overall inference speed of LLMs. In this paper, we propose an Adaptive Serial-Parallel Decoding (ASPD), which addresses two core challenges: automated construction of parallelizable data and efficient parallel decoding mechanism. More specifically, we introduce a non-invasive pipeline that automatically extracts and validates parallelizable structures from the responses of autoregressive models. To empower efficient adaptive serial-parallel decoding, we implement a Hybrid Decoding Engine which enables seamless transitions between serial and parallel decoding modes while maintaining a reusable KV cache, maximizing computational efficiency. Extensive evaluations across General Tasks, Retrieval-Augmented Generation, Mathematical Reasoning, demonstrate that ASPD achieves unprecedented performance in both effectiveness and efficiency. Notably, on Vicuna Bench, our method achieves up to 3.19x speedup (1.85x on average) while maintaining response quality within 1% difference compared to autoregressive models, realizing significant acceleration without compromising generation quality. Our framework sets a groundbreaking benchmark for efficient LLM parallel inference, paving the way for its deployment in latency-sensitive applications such as AI-powered customer service bots and answer retrieval engines.

cs.CL↗