arXiv subjects

Yu Yang

Publications and source records attributed to Yu Yang.

At least 19 recordsLinked to original sources

arXiv2026

Full Inseparability and Genuine Multipartite Entanglement Coincide for Finite-Mode Gaussian States

For general mixed states, entanglement across every bipartition need not imply genuine multipartite entanglement (GME), because a biseparable decomposition may switch the separable cut from term to term. We prove that this convex ambiguity disappears for Gaussian states of finitely many bosonic modes. More generally, for any finite family of partitions, a Gaussian density operator in the trace-norm-closed convex class generated by states separable across those partitions is already separable across one fixed partition in the family. Only the target is Gaussian; a valid decomposition may be continuous and may contain arbitrary non-Gaussian states. Thus full inseparability and GME coincide, Gaussian k-separability and k-producibility reduce to fixed-partition tests, and party-wise tensor powers cannot activate GME from a biseparable Gaussian state. The proof combines a spectral selector with a holomorphic rigidity argument that converts one product vector in the square-root range of a Gaussian state into a block-local covariance certificate. The result shows that partition mixing, a generic mixed-state mechanism, adds no new exact finite-mode Gaussian states.

quant-ph↗

arXiv2026

Reflection Steering: Disentangling Reflection from Reasoning in Activation Space for Token-Efficient Inference

Large reasoning models often produce reasoning traces with verification, revision, and backtracking. When reflection merely re-checks established results, it wastes reasoning tokens and increases latency. Most existing reflection steering methods add a label-derived mean-difference direction across preset layers, but its entanglement with reasoning and length signals destabilizes the accuracy-efficiency trade-off. In this paper, we propose Reflection Steering, a training-free framework for controlling reflection-associated computation within LLMs by disentangling reflection-related activations from general reasoning. Specifically, we contrast reflective and non-reflective hidden states at each LLM layer, denoise the resulting reflection directions with PCA, and orthogonalize them against general-reasoning directions. To limit downstream amplification from early-layer interventions, we calibrate each layer across multiple intervention strengths on a small set, retain only stable layers, and apply bounded projection removal to their residual-stream activations. We conduct extensive experiments across two public benchmarks and three open-weight LLMs against state-of-the-art activation-steering baselines. Results show that Reflection Steering reduces reasoning tokens by 16.9% on average across six matched settings. Besides, our method further introduces a bounded reflection intervention-strength parameter $\alpha$, enabling deployment-time adjustment to balance token savings, accuracy, and generation stability.

cs.LG↗

arXiv2026

Evidence-RL: Towards Evidence-intensive Visual Reasoning

Vision-Language Models (VLMs) should answer from concrete image evidence rather than language priors, dataset shortcuts, or irrelevant visual context. Existing perception-aware post-training methods encourage image use through global perturbations or attention proxies, but they do not test whether a sampled answer causally depends on the local evidence that supports it. We propose Counterfactual Evidence Disentanglement (CED), a training-time evidence audit for VLM grounding. For each response, CED neutralizes an object-centric Evidence Region and compares the resulting support drop against matched non-evidence Regions. We combine this signal with answer correctness inside GRPO, rewarding correct answers that rely on the evidence path rather than shortcut or nuisance paths. CED uses weak object-level proposals, requires no question-specific evidence annotations, and adds no inference-time overhead. Across nine public benchmarks and four backbones, CED outperforms prior RL-based post-training methods, with targeted analyses verifying its object-centric signal.

cs.CV↗

arXiv2026

Quo Vadis, World Modeling?

Continually improving agents require dynamic interaction feedback beyond static supervision, yet direct real-environment interaction is costly, slow, unsafe, and hard to parallelize. World modeling offers a natural intermediate proxy that allows agents to query lower-cost, more controllable feedback before committing to real actions. Classical world models instantiate this proxy primarily through future physical-state prediction, a formulation useful yet narrow for agents that require actionable feedback beyond raw state transitions. In this work, we conceptualize Agent-Centric Interactive World Proxies, shifting the fundamental paradigm from physical state transitions to agent-usable information transitions, such as execution outcomes, retrieved experiences or skills, and verification signals, broadening the scope of world modeling to provide versatile feedback for continually improving agents. To systematically map this design space, we organize world proxies into six functional forms based on their feedback modalities: dynamics, spatial, execution, memory/experience, skill, and reward/verification proxies, which together characterize the primary ways world modeling serves agent improvement. We further analyze how these proxies empower agents across three progressive levels: L.1 Inference-Time Guidance, where proxy outputs enrich in-context information for superior decisions; L.2 Training-Time Optimization, where proxy outputs yield rewards, critiques, or synthetic rollouts for policy learning; and L.3 Agent-Proxy Co-Evolution, where real-environment evidence continuously updates both the proxy and the agent for co-evolution. Ultimately, this work recasts world modeling into an agent-centric paradigm, establishing a roadmap for building world proxies that empower agents to plan better, learn faster, and evolve continually.

cs.CV↗

arXiv2026

Distill Where You Fail: Recovering Learning Signals of Negative RL-Groups from Adaptive Teacher Guidance

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models (LLMs). While Group Relative Policy Optimization (GRPO) is widely adopted, it suffers from sparse reward signals and loses gradients entirely when all responses within a group receive identical rewards. On-policy distillation (OPD) offers a natural remedy by providing dense, token-level supervision from a teacher model. However, naively combining GRPO with OPD leads to degraded performance, due to three underlying causes: not all samples benefit from distillation; fitting too quickly to the teacher undermines the exploratory capacity of RL; and OPD's advantages are asymmetric, suppressing most tokens. To address these challenges, we propose RSTG (Recovering Learning Signals via Adaptive Teacher Guidance), which applies distillation selectively and precisely where it matters most. At the sample level, OPD is restricted to negative zero-variance prompts with each sample weighted by the teacher's confidence score. At the token level, distillation targets only tokens with high student entropy or large teacher-student divergence. We further augment training with SFT on correct trajectories generated by the teacher model, injecting positive gradient signals where RL yields none. Experiments demonstrate that RSTG substantially outperforms naive GRPO+OPD by +4.02% on math and +3.05% on code.

cs.CL↗

arXiv2026

Observation of Moir\'e Time Crystal in Floquet-driven Rydberg Atomic Gases

A Moir\'e time crystal is a non-equilibrium quantum phase emerging from the coherent interference of two distinct frequencies, at least one being the intrinsic oscillation of a symmetry-broken time crystal. Its hallmark is an ultra-long beat period, reflecting a time-domain mapping of the Moir\'e fringes that arise from mismatched spatial lattices. However, to date, no experimental realization of such a Moir\'e time crystal has been reported. In this work, by applying a bichromatic driving field with two distinct frequencies, we demonstrate that the interplay between long-range Rydberg interactions and dissipation gives rise to a unique comb-like Moir\'e pattern characterized by a beat-note comb, which superimposes subharmonic periodicity and fundamental frequencies. This Moir\'e pattern formed by two mismatched drives is staggered in the spectrum as the frequency of one driver changes. We experimentally map the phase diagram of the system and identify a robust region where the Moir\'e temporal order persists against perturbations in laser detuning. The reported Moir\'e time crystal not only provides a controllable platform for exploring emergent slow-fast dynamics and synthetic space-time symmetries but also opens avenues for engineering complex temporal order in driven quantum many-body systems.

cond-mat.quant-gas↗

arXiv2026

MemHarness: Memory Is Reconstructed, Not Replayed

Retrieving past experiences has become a common strategy to enhance large language model agents. However, most existing memory-augmented agents treat retrieved experiences as static records to be replayed verbatim, injecting them into the context regardless of whether they align with the agent's current situation. This ``replay'' paradigm ignores the gap between the abstract, general nature of stored experience and the concrete, ever-changing states encountered at decision time, frequently causing negative transfer. In contrast, humans rarely recall past experiences verbatim; instead, they reorganize and adapt retrieved memories to fit the present context. Inspired by this, we propose MemHarness, a framework that equips LLM agents to actively harness and reconstruct past experiences based on the present context. At each decision step, a unified policy model critiques and reconstructs the retrieved experience conditioned on the current state, producing context-grounded guidance before acting. This reconstructive ability emerges naturally through end-to-end training with GRPO. Experiments on ALFWorld and WebShop show that MemHarness substantially outperforms pure RL and static memory-augmented baselines, demonstrating strong robustness in out-of-distribution (OOD) scenarios. Furthermore, our analyses reveal that this reconstruction objective not only prevents negative transfer but also serves as latent guidance during training, fundamentally improving the agent's intrinsic reasoning capabilities.

cs.AI↗

arXiv2026

Full-Pipeline Inference Optimization for MiMo-V2.5 Series: Pushing Hybrid SWA Efficiency to the Limit

We present a full-pipeline inference optimization for the MiMo-V2.5 model family, which combines Hybrid Sliding Window Attention (Hybrid SWA), sparse Mixture-of-Experts (MoE), and multimodal encoders. While Hybrid SWA can ideally reduce both attention compute and KVCache storage significantly compared to Full Attention, realizing these gains in production requires substantial engineering effort. We systematically optimize the KVCache system with layerwise prefetch, SWA-aware prefix cache trees, and specialized placement strategies, achieving strict $O(W)$ SWA storage and high cache hit rates. We further build GCache, a high-performance distributed cache infrastructure with RDMA-optimized networking, and develop a KVCache-affinity router to reduce computation while preserving load balancing. We also optimize for multimodal inputs, including GPU image preprocessing, parallel video decoding, and multimodal cache sharing. Together, these optimizations constitute the first large-scale LLM serving system in production that efficiently covers the Hybrid SWA + MoE + multimodal composite architecture.

cs.AR↗

arXiv2026

Proxy OPD: On-Policy Distillation with Transferable Relative Proxy Update

Post-training for large language models typically couples policy exploration with model optimization, hindering the reuse of high-reward behaviors from policy exploration. While on-policy distillation alleviates this by consolidating independently optimized experts, its reliance on matching absolute expert distributions can yield suboptimal supervision, especially when the target model possesses a different prior or already surpasses the expert's capabilities. To alleviate this, we introduce Proxy OPD (P-OPD), an asynchronous post-training framework that transfers reward-induced policy improvements rather than absolute policy distributions. P-OPD first optimizes a proxy policy via reward feedback. It then extracts the relative distributional changes between the proxy's initial and optimized states, transferring these directional updates through the target model's own on-policy trajectories while retaining the target policy as the reference. This decoupled formulation requires the proxy to provide merely a useful direction of improvement rather than superior absolute capability, enabling update signals from older or weaker proxies to remain highly effective. Systematic experiments on Qwen3-family models across mathematical reasoning and code generation demonstrate that P-OPD consistently enhances already strong target models. Furthermore, transfer intensity can be dynamically modulated through signal scaling, making the extracted update signals seamlessly reusable across diverse model variants and training configurations. These results establish relative policy updates as highly reusable, adjustable assets for scalable, reward-based post-training.

cs.LG↗

arXiv2026

Purifying one-neutron removal as a probe of single-particle strength

One-neutron removal reactions exhibit a strong proton-neutron asymmetry dependence in the inclusive reduction factor $R_s$, a long-standing issue that has been discussed in terms of both possible intrinsic isospin dependence of single-particle strength and reaction-mechanism effects. We address this issue by reframing inclusive removal as a coupled fast-dynamics and deexcitation process, and by validating this transport-deexcitation chain against a global, mutually constraining data set. Confronting 73 one-neutron removal cross sections and 28 residue parallel-momentum distributions with isospin-dependent quantum molecular dynamics followed by GEMINI evaporation shows that the apparent $R_s$-$\Delta S$ trend is correlated with evaporation feeding and evaporation loss. By subtracting the feeding contribution and correcting for the loss component in the measured cross sections, we construct a purified reduction factor $R_{\rm dir}$, that more closely reflects single-particle strength than the inclusive $R_s$. The resulting $R_{\rm dir}$ exhibits a much weaker $\Delta S$ dependence within current uncertainties, consistent with the weak isospin-asymmetry dependence observed in nucleon-transfer and quasifree-knockout systematics.

nucl-th↗

arXiv2026

SAERec: Constructing Fine-grained Interpretable Intents Priors via Sparse Autoencoders for Recommendation

Intent-based recommender systems have gained significant attention for improving accuracy and interpretability by modeling the underlying motivations behind user behaviors. Most existing models derive intents directly from user sequences via clustering or prototype learning. However, they are sensitive to sequence quality, require presetting the number of intents, and lack explicit semantic grounding. These issues lead to an incomplete and coarse intent set and limit the effectiveness of recommendation. In this paper, we propose the Sparse Autoencoder for intent-based recommendation (SAERec), a novel recommender that automatically constructs a fine-grained and interpretable intent space from a textual corpus to guide recommendation. Rather than treating texts as side signals, SAERec leverages them as high information density evidence for intent construction. Specifically, we first extract a comprehensive set of fine-grained interpretable intents from the latent space of large language models (LLMs) by using a sparse autoencoder (SAE) to disentangle and interpret text embeddings, which isolates intent-related semantics from textual noise. Then, for each user, we retrieve relevant intents from this set as priors to guide recommendation. It contains personal intents matching a user's current interests and public intents capturing general item patterns shared across users (e.g., quality, price). Finally, to integrate retrieved intents into sequence modeling, we propose a multi-branch attention mechanism that captures temporal dependencies and injects both personal and public intent signals, followed by an adaptive fusion layer to construct the final user representation for recommendation. Extensive experiments on public datasets demonstrate the superiority of SAERec, consistently outperforming state-of-the-art baselines while providing human-understandable explanations.

cs.IR↗

arXiv2026

ComAct: Reframing Professional Software Manipulation via COM-as-Action Paradigm

Existing computer-use agents remain fundamentally limited in professional software manipulation: GUI-based agents suffer from fragile visual grounding and long-horizon error accumulation, while API-basedapproaches struggle with heterogeneous protocols and inaccessible commercial interfaces. In this work,we identify the Component Object Model (COM) as a unified executable abstraction, proposing COM-as-Action: a new paradigm that reframes professional software interaction as deterministic program synthesisrather than sequential visual control. To validate this paradigm in the most demanding environments, weintroduce ComCADBench, the first benchmark for agents operating real industrial CAD software. Ourexperiments reveal a substantial paradigm gap: frontier proprietary models achieve near-zero successunder GUI-based interaction, whereas COM-based execution yields substantial immediate gains. Tobridge the remaining gap between syntactic correctness and geometric accuracy, we develop ComActor, aself-correcting agent trained through a progressive three-stage framework, alongside ComForge, a scalableplatform for large-scale training in Windows containers. Extensive experiments show that ComActorachieves state-of-the-art performance on ComCADBench, with strong resilience in long-horizon taskswhere baselines collapse, and generalizes to external CAD benchmark.

cs.SE↗

arXiv2026

IterCAD: An Iterative Multimodal Agent for Visually-Grounded CAD Generation and Editing

Computer-Aided Design is pivotal in modern manufacturing, yet existing automated methods predominantly rely on open-loop, one-shot generation, creating a mismatch with iterative real-world practices. In this paper, we present IterCAD, a unified multimodal agent framework for closed-loop, interactive CAD generation and editing. We formulate the task as a multi-turn interaction between a multimodal agent and an executable CAD sandbox, covering three tasks: Drawing-to-Code, Text-to-Code, and Interactive Editing. To support this, we develop a data synthesis pipeline incorporating advanced industrial manufacturing features to generate standard-compliant multi-view engineering drawings, complex code-editing tasks, and high-fidelity interaction trajectories. We optimize the agent via progressive SFT followed by geometry-aware reinforcement learning with viable-prefix masking to enhance code executability and geometric fidelity. Finally, we introduce the IterCAD-Bench evaluation suite and propose the Chamfer Distance Tolerance-Recall (CD-TR) curve alongside its AUC-TR metric, establishing a survivor-bias-free standard that unifies code validity and geometric precision. Extensive experiments demonstrate that IterCAD achieves highly competitive performance across multiple benchmarks, significantly outperforming existing approaches in both code executability and geometric precision, while exhibiting superior capabilities in closed-loop iterative refinement.

cs.AI↗

arXiv2026

Reasoning or Memorization? Direction-Aware Diversity Exploration in LLM Reinforcement Learning

Reinforcement learning has become a key paradigm for eliciting reasoning abilities in large language models, where exploration is crucial for discovering effective solution trajectories. Existing exploration methods typically encourage diversity in semantic or gradient spaces, without distinguishing what drives this diversity. A trajectory may appear novel because it follows a new reasoning process, or because it varies memorized patterns and shortcuts. Rewarding both cases equally may steer exploration toward memorization rather than genuine reasoning improvement. In this paper, we propose DiRL, a Direction-Aware Reinforcement Learning framework that anchors exploration to an internal reasoning-memorization direction of the policy. Specifically, DiRL extracts this direction from model representations, constructs direction-weighted gradient features to characterize rollout updates, and shapes rewards to amplify reasoning-aligned exploration while suppressing memorization-aligned variations. DiRL integrates seamlessly into standard Group Relative Policy Optimization (GRPO). Extensive experiments on mathematical and general reasoning benchmarks demonstrate the effectiveness of DiRL, showing significant improvements over various existing exploration methods.

cs.AI↗

arXiv2026

Towards Event-Robust Acoustic Scene Classification

This paper introduces the Event-Shifted Acoustic Scene (ESAS) dataset, a novel benchmark for evaluating the robustness of Acoustic Scene Classification (ASC) systems against unknown sound events. Existing ASC datasets typically contain recordings of clean and consistent audio, while real-world environments often include diverse and unexpected sound events. To bridge this gap, ESAS simulates real-world acoustic variability by injecting foreground sound events into background scenes with the assistance of large language models. In this work, we present the construction methodology, dataset statistics, and evaluation protocols. Furthermore, a comprehensive evaluation of state-of-the-art ASC systems is conducted using the ESAS benchmark. Experimental results reveal that existing ASC models suffer significant performance degradation when facing the event-shift challenge. The introduction of the ESAS dataset aims to drive future research toward event-robust ASC.

cs.SD↗

arXiv2026

FinEvolveBench: A Benchmark for Self-Evolving Agents on Low-Repetition Tasks with Implicit Rewards

Experience-based self-evolution enables language-model agents to improve their behavior by accumulating and updating experience at test time, yet existing evaluations often assume recurring task patterns and explicit success signals. We introduce \textsc{FinEvolveBench}, a benchmark for self-evolving agents on low-repetition tasks with implicit rewards. The benchmark reconstructs a daily financial information stream over 31 Chinese A-share industry indices and aligns 177,324 public news articles with market observations. Researchers can define prediction horizons over this stream; we evaluate predictive market-sentiment factors against delayed market-adjusted returns after 10, 20, and 40 trading days. Unlike static benchmarks that score each prediction independently, \textsc{FinEvolveBench} interleaves new decisions with delayed outcomes from earlier ones, testing whether agents can convert noisy real-world feedback into reusable experience at test time. Experiments with two backbone models show that the evaluated general-purpose memory systems do not consistently outperform the no-experience pipeline. A matched ablation further shows that feedback-driven utility updates help at shorter horizons on one backbone but hurt in most settings on the other. Together, these results position \textsc{FinEvolveBench} as a diagnostic testbed for experience-based self-evolution under noisy, delayed, and outcome-level feedback.

cs.CL↗

arXiv2026

Cross-Domain Energy-Guided Diffusion Generation for Off-Dynamics Reinforcement Learning

Off-dynamics offline reinforcement learning seeks to learn a target-domain policy from a large source dataset and a limited target dataset under mismatched transition dynamics. Existing approaches such as reward augmentation and data filtering are constrained to the source dataset and cannot synthesize new target behavior to improve coverage beyond the collected source trajectories. While recent model-based methods attempt to address this by learning target-aware dynamics, the generated experience is constructed only at the transition level, which leads to accumulated errors over long horizons. These limitations necessitate a shift toward trajectory-level generation for off-dynamics offline RL. We propose CEDGE, a Cross-domain Energy-guided Diffusion GEneration framework. CEDGE trains a trajectory diffusion model on source-domain trajectories and adapts the generated samples to the target domain through energy guidance. This guidance is derived by minimizing the distribution mismatch between the source and desired target-domain trajectories and is decomposed into return, domain, and behavior energy components. The resulting energy-guided trajectories are useful both for direct planning and as synthetic data for policy learning. Since target adaptation is achieved via energy guidance rather than retraining the diffusion model, CEDGE can be efficiently adapted to new target dynamics compared to previous methods. Experiments on the ODRL benchmark demonstrate that trajectory-level energy-guided generation improves diffusion planning under dynamics shifts and produces synthetic data that improves downstream target policy learning.

cs.LG↗

arXiv2026

Inference-Time Alignment of Diffusion Models via Trust-Region Iterative Twisted Sequential Monte Carlo

We study inference-time alignment for diffusion-based generative models, aiming to steer a base model toward high-reward outputs without updating its weights. Recent Sequential Monte Carlo (SMC)-based steering methods approximate reward-tilted target distributions in a principled way, but their proposals remain largely tied to the base sampler. Since reward information is mainly used after propagation through particle reweighting and resampling, these methods can require large particle budgets and suffer from weight degeneracy and high-variance estimates. One way to reduce variance and improve particle efficiency is to iteratively learn twisting functions that provide look-ahead guidance, as in twisted SMC. However, existing learnable twisting methods are developed mainly for classical sequential inference and can be unstable when applied to diffusion-based alignment with high-dimensional state spaces and terminal, noisy, or black-box rewards. We propose Trust-Region Iterative Twisted Sequential Monte Carlo (TRI-TSMC), a trust-region framework for learning twisting functions in SMC-based inference-time alignment. Each iteration computes an exact KL-constrained update in path space, which admits a closed-form solution by tempered importance reweighting, and projects this target back to the parameterized twisted family by weighted maximum likelihood. Theoretically, we formalize the value-function interpretation of the optimal twisting function and show that it yields a zero-variance sampler. We prove that the trust-region update follows an escort path toward the target distribution, that the weighted maximum-likelihood update is a forward-KL projection, and that the path reduces residual importance-weight variance. Empirically, TRI-TSMC improves primary alignment objectives on discrete diffusion text generation and text-to-image generation under matched inference-time budgets.

cs.LG↗