arXiv subjects

Yu Yang

Publications and source records attributed to Yu Yang.

At least 19 recordsLinked to original sources

arXiv2026

Language Model Augmented Decision Transformer for Multi-Task Offline Reinforcement Learning

Offline multi-task reinforcement learning (MTRL) aims to learn a unified policy from heterogeneous offline datasets that generalizes across tasks. However, existing offline MTRL methods often struggle with two fundamental issues: severe per-task data scarcity and limited cross-task generalization. In this paper, we propose Language-Model Augmented Decision Transformer (LaMA-DT), a unified sequence-modeling framework that leverages a single large language model (LLM) for both trajectory augmentation and multi-task policy learning. LaMA-DT uses natural-language task descriptions to guide task-aligned trajectory augmentation through a set of supervised generation and alignment objectives, and incorporates a scoring-and-filtering mechanism to retain high-quality augmented trajectories. The same LLM is then repurposed as a language-conditioned Decision Transformer and fine-tuned on a mixture of real and augmented data. Experiments on Meta-World and MuJoCo multi-task benchmarks demonstrate that LaMA-DT consistently outperforms competitive baselines, particularly in low-data regimes, showing that language-guided data augmentation substantially improves cross-task generalization.

cs.CL↗

arXiv2026

RedCodeAgent: Automatic Red-teaming Agent against Diverse Code Agents

Code agents have gained widespread adoption due to their strong code generation capabilities and integration with code interpreters, enabling dynamic execution, debugging, and interactive programming capabilities. While these advancements have streamlined complex workflows, they have also introduced critical safety and security risks. Current static safety benchmarks and red-teaming tools are inadequate for identifying emerging real-world risky scenarios, as they fail to cover certain boundary conditions, such as the combined effects of different jailbreak tools. In this work, we propose RedCodeAgent, the first automated red-teaming agent designed to systematically uncover vulnerabilities in diverse code agents. With an adaptive memory module, RedCodeAgent can leverage existing jailbreak knowledge, dynamically select the most effective red-teaming tools and tool combinations in a tailored toolbox for a given input query, thus identifying vulnerabilities that might otherwise be overlooked. For reliable evaluation, we develop simulated sandbox environments to additionally evaluate the execution results of code agents, mitigating potential biases of LLM-based judges that only rely on static code. Through extensive evaluations across multiple state-of-the-art code agents, diverse risky scenarios, and various programming languages, RedCodeAgent consistently outperforms existing red-teaming methods, achieving higher attack success rates and lower rejection rates with high efficiency. We further validate RedCodeAgent on real-world code assistants, e.g., Cursor and Codeium, exposing previously unidentified security risks. By automating and optimizing red-teaming processes, RedCodeAgent enables scalable, adaptive, and effective safety assessments of code agents. Our code and data are available at https://github.com/1mocat/RedCodeAgent.

cs.SE↗

arXiv2026

SAG-Sep: Sparse Augmented Graphs and Onion-Guided Search for Rounded Capacity Cut Separation

Rounded capacity cuts (RCCs) are among the most effective cuts for capacitated vehicle-routing relaxations, but exact separation is computationally prohibitive at scale. Existing heuristic and neural separators can be fast, but often fail to produce sufficiently many effective cuts within practical separation budgets. We introduce SAG-Sep, a separator that scores node membership in RCC-inducing subsets across vehicle-count levels from a fractional LP solution in a single encoder forward pass, avoiding the iterative predict-and-coarsen inference used by NeuralSEP-style separators. SAG-Sep encodes the LP solution as a Sparse Augmented Graph (SAG), adding flow-informed multi-hop edges that expose long-range routing structure without densifying attention. Beyond LP-aware sparse encoding, we observe that high-value exact RCCs are typically nested, organizing into onion-like subset families, and SAG-Sep's node scores track subset nesting depth. This motivates Onion-Sweep (OS), a training-free decoder that ranks customers by their scores, evaluates the RCC violation of every prefix of the ranking in one vectorized pass, and emits the most violated prefix of each replica together with a cross-replica family of nested proposals, converting independent node scores into structurally coordinated RCC candidates without a tuned threshold or probing band. On the NeuralSEP benchmark with N=1000 customer instances, SAG-Sep with OS reduces the average root gap by 17.7% relative to the strongest NeuralSEP baseline, while the single-pass SAG-Sep separator achieves a roughly 30-fold per-iteration neural separation speedup. Our findings suggest that LP-aware sparse encoding and the nested structure of high-value cuts are structural patterns worth leveraging in neural separation more broadly.

cs.LG↗

arXiv2026

CRISP: Co-designing Pruning and Scheduling for Efficient MoE Inference on Edge Servers

Mixture-of-Experts (MoE) inference on memory-constrained edge platforms is often dominated by expert-weight movement rather than computation. Although each token activates only a small subset of experts, routing is input-dependent, so all experts must remain accessible and memory movement dominates latency. Existing approaches such as uniform pruning and expert dropping overlook two properties of MoE execution: expert combinations are highly skewed in frequency, and experts contribute asymmetrically within each combination. We present CRISP, a combination-aware framework that co-designs pruning and runtime scheduling for efficient MoE serving on memory-constrained edge platforms. CRISP reorders neurons by importance so one stored expert can support multiple execution widths, assigns differentiated execution to hot, warm, and cold combinations based on frequency and expert role, and exploits heterogeneous widths to overlap expert loading with computation. Across Mixtral-8x7B and DeepSeek-MoE-16B, CRISP improves throughput by up to $2.1\times$ at the same memory budget while limiting accuracy loss to 3.6\% relative to full-width inference. Our results highlight that efficient edge MoE serving requires jointly optimizing which expert capacity is preserved and how expert execution is scheduled.

cs.LG↗

arXiv2026

Nodal Inertia Pricing for RoCoF Security with Integration of Synchronous and Virtual Resources

Frequency regulation is becoming a more significant challenge in the undergoing low-carbon transition of modern power systems, which is heavily rely on system inertia provision. In the context of market-based energy allocation, reasonable valuation of the system inertia is crucial to its procurement, namely a worthy price should be determined for inertia providers. However, current research on the value of inertia remains insufficient, particularly during the initial phase immediately following a contingency, a critical period when the system primarily relies on the inherent inertia from conventional units and pre-engaged virtual inertia resources. Furthermore, it is found in existing studies that the Center of Inertia (COI) models exhibit significant deviations from actual physical conditions. For market-oriented reforms that explore individual differences, COI models also fail to capture the heterogeneous contributions of inertia at different nodes to system security. Given these circumstances, this paper proposes a nodal inertia pricing mechanism via a day-ahead unit commitment and inertia co-optimization model with nodal resolution. Following the proposed pricing scheme, the day- or hour-ahead inertia provision is monetized and promoted, providing regional guidance to units and devices with virtual inertia provision, which is further verified by simulation results of the South East Australia Power System.

eess.SY↗

arXiv2026

Playsemble: Learning Low-Level Programming Through Interactive Games

Teaching assembly programming is a fundamental component of undergraduate computer science education, yet many students struggle with its abstract and low-level concepts. Existing learning tools, such as simulators and visualisers, support understanding by exposing machine states. However, they often limit students to passive observation and provide few opportunities for meaningful interaction. To address these limitations, we introduce Playsemble, a gamified learning system that transforms assembly instructions into interactive, game-like tasks in which students control Pac-Man to collect items, avoid ghosts, and reach targets. Playsemble integrates a code editor, a CPU emulator, and visual debugging tools within a browser-based environment, allowing students to work offline without installation or configuration. It also provides immediate formative feedback enhanced by large language models. We deployed Playsemble in an undergraduate computer architecture course with 107 students. The course featured a sequence of assignments of increasing complexity, covering core concepts such as register and memory manipulation, control structures including loops and conditionals, and arithmetic operations. Our findings suggest that Playsemble promotes active experimentation, sustained engagement, and deeper conceptual understanding through meaningful game-based learning experiences.

cs.CY↗

arXiv2026

Orbia: Do Generated Videos Form a Consistent 3D World?

Video generators serve as interactive worlds for exploration and simulation, yet realistic and visually appealing frames do not necessarily depict a consistent 3D world. Existing benchmarks assess selected geometric properties or use vision-language models to judge memory and consistency, leaving systematic 3D evaluation underexplored. We introduce Orbia, the first benchmark dedicated to a systematic and comprehensive evaluation of 3D consistency in interactive world models. Through controlled exploration and revisit, we organize the evaluation into three dimensions: input preservation, generated-content persistence, and 3D self-consistency. For input preservation, registered RGB-D references and annotated anchors support appearance, geometry, and spatial stability measurements. For generated-content persistence, we compare revisits to regions absent from the input image with their first generated observations to test whether models preserve what they create. For 3D self-consistency, cross-view depth and feature correspondence, regional scale agreement, and held-out Gaussian rendering test whether generated views describe the same 3D scene. Minute-long exploration and revisits further test long-term consistency. Our construction and annotation pipeline brings together public datasets, self-captured scenes, and simulation, yielding 720 instances across indoor and outdoor environments and six exploration-and-revisit protocols. Evaluation spans 27 text-, action-, and camera-conditioned models through native-interface adapters. Our Orbia benchmark provides a systematic basis for diagnosing 3D inconsistency and developing more persistent video world models.

cs.CV↗

arXiv2026

Multilevel Neural Corrections with Adaptive Sampling for Partial Differential Equations

We propose a multi-level framework for deep neural PDE solvers that integrates adaptive sampling with additive neural corrections. At each level, an adaptive sampling distribution is constructed from normalized residual-gradient indicators inherited from the preceding levels, thereby concentrating collocation points in regions where the current approximation exhibits large local errors or reduced regularity. The neural approximations obtained at previous levels are incorporated into a successive correction problem, leading to a hierarchical refinement of the numerical solution. Under suitable assumptions, we establish rigorous approximation results and derive error bounds for the proposed multi-level framework, which quantify the cumulative effect of the successive neural corrections across levels. Numerical experiments on representative PDEs with steep gradients and low-regularity solutions demonstrate the effectiveness of the adaptive multi-level strategy.

math.NA↗

arXiv2026

VITAL: From Visual Motion to Precise Action

World-action models aim to turn visual anticipation into driving decisions, yet typically train action experts on trajectory representations defined independently of visual dynamics. This motivates deriving action targets directly from visual motion. We first examine a simpler setting in which trajectories are reconstructed from recorded future images. Even without the need for future prediction, reconstruction captures the dominant maneuver but leaves centimeter-scale errors. We therefore introduce VITAL (Vision-Induced Tokens for Action Learning), an action tokenizer that couples a visually derived motion scaffold with a compact residual encoding the trajectory error left by that scaffold. This representation makes visual motion the principal prediction target and gives small metric corrections a separate, normalized learning scale. Using flow matching, the action expert jointly predicts both components from the current image, ego state, and navigation command, without generating future images at inference. The same residual coordinates support bounded reinforcement post-training that refines execution while keeping the predicted visual scaffold fixed. With only one action denoising step, VITAL achieves 90.8 PDMS on NAVSIM v1 navtest, within 0.13 points of its ten-step score at 3.29x the end-to-end inference speed. Without retraining, it achieves 91.1 EPDMS on NAVSIM v2 navtest and 38.6 on navhard. Project page: https://worldbench.github.io/vital

cs.CV↗

arXiv2026

Imperfect Is Perfect: Learned Column Prefilling for Branch-Price-and-Cut

Column generation folklore warns that initial columns hurt: a warm start may bias the dual and slow the root of branch-price-and-cut. We show the opposite once the columns are imperfect in the right way. Preloading the support of the optimal root LP, the best columns one could know in advance, fails to finish on half of our hardest instances; a pool of ordinary routes that need not belong to any optimal solution, held outside the master and priced at the current smoothed dual, speeds the root up. Because the columns need not be optimal, producing them becomes a learning task: a neural edge scorer guides a constructive generator that builds the pool in seconds. A second network predicts the initial center of Wentges smoothing from the instance file, and with it the direction of the smoothed pricing queries; the solver's own updates run unchanged. With prefill and center in place, exact pricing dominates root pricing time, so we rebuild bucket-graph labeling around its dominance test for the GPU and certify on the CPU. Inside RouteOpt, the state-of-the-art exact solver for capacitated vehicle routing, the pipeline reaches the same certified root bound in 13.2% of vanilla RouteOpt's root pricing time over 95 instances with 300 to 1000 customers, prefill alone in 26.1%, and 17.5 times faster in the best case. In an exact branch-price-and-cut solver for unrelated parallel machine scheduling, prefill with smoothing and the learned center certifies all 25 benchmark instances, 12 times faster than the solver's own column generation, which certifies 20 of them.

cs.LG↗

arXiv2026

Vision: "What Don't I Know About User's Preferences?" How Uncertainty Estimation Improves Proactive Assistant Alignment

Proactive Smart Assistants (PSAs) increasingly use general knowledge, learned patterns from group-level data, patterns from individual user's data, and recent sensor observations to anticipate which action a user may prefer. Existing systems typically select actions from predicted user preference likelihoods, but these likelihoods alone do not reveal how reliable the prediction is, which evidence supports it, or why ambiguity remains. In this vision paper, we argue that uncertainty estimation can provide this missing information and substantially change appropriate downstream behavior. Using the example of a home assistant, we introduce the questions that uncertainty estimation aims to answer, i.e. how sure is the assistant about the predicted preference likelihood, whether the confidence come from group-level assumptions or individual's data, and whether likelihood ambiguity reflect the user's intrinsic preference variability or insufficient data for the particular situation. We further introduce uncertainty estimation applications in neighboring domains, such as human trajectory prediction, personalized activity recognition, information-collection, and model-based control. Finally, we draw on insights from neighboring domains to envision how future uncertainty-aware proactive assistants may address key challenges in proactive interaction.

cs.AI↗

arXiv2026

Rethinking Relative Policy Differences Transfer of On-Policy Distillation

On-policy distillation (OPD) trains a target policy on its own rollouts using token-level supervision from a stronger teacher, but typically relies on the teacher being substantially more capable than the target. Recent work relaxes this requirement by transferring the relative policy difference between an expert and a reference policy rather than the expert's absolute distribution. Yet it remains unclear which differences provide useful supervision and which targets can benefit from them. We develop a unified view in which the log-probability ratio between a proxy expert and a proxy reference defines an implicit reward for KL-regularized policy optimization. This formulation induces an on-policy update for the target, recovers conventional OPD as a special case, and enables any-to-any transfer across models that differ in scale, training stage or generation. Our systematic experiments yield three findings. First, a stronger expert is not necessarily a better source of supervision: differences induced by task-specific training, including reinforcement learning and supervised fine-tuning, transfer consistently, whereas those arising from model scale alone transfer weakly or not at all. Second, the expert and reference serve complementary roles: the expert provides the capability to be transferred, while the reference filters out shared behavior and isolates the relevant residual. Third, even a useful difference does not transfer to every target: success requires the induced update to move the target toward the expert and generalize beyond sampled rollouts. Raw policy distance does not predict transferability, whereas an appropriate reference can rescue otherwise unsuccessful transfer. These results characterize when relative policy differences provide effective supervision for OPD.

cs.LG↗

arXiv2026

Characterizing chordal $\llcorner$-EPG graphs via admissible clique tree orientations

A $k$-bend path is a non-self-intersecting polyline that lies on a grid and consists of at most $k+1$ axis-parallel line segments. An $\llcorner$-EPG graph is a graph whose vertices can be represented by 1-bend paths on a grid, where each path is either $\llcorner$, $\shortmid$, or $\text{-}$, such that two vertices are adjacent if and only if the corresponding paths share at least one grid edge. Characterizing chordal $\llcorner$-EPG graphs was explicitly posed as an open problem by Cameron, Chaplick, and Hoàng. We resolve this by giving two equivalent characterizations of connected chordal $\llcorner$-EPG graphs. The first assigns to every connected chordal $\llcorner$-EPG graph an ordered clique partition tree of its vertex set, where the underlying tree is the bipartite incidence graph of the horizontal and vertical edge-intersection components. The second is stated purely in terms of maximal cliques: a connected chordal graph is $\llcorner$-EPG if and only if some clique tree admits an admissible partial orientation. A reduction lemma replaces local clique labels by maximal cliques, linking the two characterizations. For a prescribed clique tree, the existence of an admissible partial orientation reduces to 2-SAT and is decided in $O(nM^2)$ time, where $n$ and $M$ are the numbers of vertices and maximal cliques. Finally, we show that strong chordality is not sufficient: we exhibit a strongly chordal graph that is a minimal forbidden induced subgraph for $\llcorner$-EPG, and a family of split graphs with a unique clique tree for which membership in $\llcorner$-EPG reduces to a neighborhood condition.

math.CO↗

arXiv2026

Resolving Two Open Problems of Planar $B_k$-CPG Recognition: $k=0,1$

A $k$-bend path is a non-self-intersecting polyline that lies on a grid and consists of at most $k+1$ axis-parallel line segments. A $B_k$-CPG graph is a graph whose vertices can be represented by pairwise interiorly disjoint $k$-bend paths on a grid such that two vertices are adjacent if and only if the corresponding grid paths touch at a grid point. We prove that recognizing planar $B_0$-CPG graphs of maximum degree 8 is NP-complete, and that recognizing planar $B_1$-CPG graphs of maximum degree 11 is NP-complete. These results settle two of the three planar recognition problems left open by Champseix, Galby, Munaro, and Ries.

math.CO↗

arXiv2026

Covertness as a resource constraint in quantum target sensing

Quantum covert target sensing aims to detect low-reflectivity targets embedded in the thermal environment while avoiding discovery by an adversary. This covertness requirement constrains the admissible probe energy, thereby imposing a trade-off between covertness and sensing performance. However, how this resource constraint determines the ultimate estimation precision of the target parameter remains largely unexplored. To fill this gap, we establish a unified framework connecting covert target discrimination with quantum-enhanced parameter estimation. We first present perturbative $ε$-covertness-induced bounds on the probe energy, which yield an admissible energy window narrowing as $\mathcal{O}(\sqrt{ε/M})$. Moreover, we introduce the covertness-constrained quantum Fisher information (CCQFI) to quantify the ultimate precision of target-reflectivity estimation and assess its attainability with specific measurements. For Gaussian-distributed coherent-state (GCS) and two-mode squeezed vacuum (TMSV) probes, their CCQFIs scale as $\mathcal{O}(ε)$ and $\mathcal{O}(M)+\mathcal{O}(\sqrt{εM})+\mathcal{O}(ε)$, respectively. In particular, in the strong covertness regime $ε\to0$, the GCS CCQFI vanishes, whereas the TMSV retains a finite $\mathcal{O}(M)$ contribution exhibiting a pronounced metrological advantage. Besides, photon counting saturates the GCS CCQFI, whereas homodyne detection achieves a higher attainability for the TMSV probe. Our work identifies covertness as a quantitative resource constraint governing the metrological limits of quantum target sensing.

quant-ph↗

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

EvoSkillRec: Skill-Genome Evolution for Recommender Architecture Discovery

Modern recommender systems advance not only by scaling data and parameters, but also by encoding task-specific inductive biases through architecture, including sparse feature interactions for click-through rate (CTR) prediction, temporal attention for sequential recommendation, and expert routing for multi-task learning. However, these biases are typically human expert designed or searched within predefined operator spaces. Although Recent LLM-driven code evolution expands this space, unconstrained edits often produce invalid or ineffective architectures, underuse established architecture design knowledge, and fail to preserve successful innovations for reuse. We introduce EvoSkillRec, a promotion-and-reuse framework for cumulative recommender architecture evolution. It first decomposes recommenders into atomic executable skills and represents architectures as typed skill genomes, with each skill equipped with input--output types, semantic annotations, and implementation code. We then evolve models with different tasks through two coupled spaces: a constrained skill--space that mutates, recombines, specializes, and reuses validated skills, and an open-ended code--space in which LLM planners and synthesizers invent new skill modules using prior evolution traces and accumulated experience. An autoresearch controller evaluates candidates, diagnoses failures, retrieves relevant skills, promotes validated innovations into the skill library, and adaptively allocates the proposal budget between the two spaces. Extensive experiments on CTR prediction, multi-task learning, and multi-domain learning, including resource-constrained co-optimization of predictive quality and model FLOPs utilization in generative ranking models, consistently demonstrate the effectiveness of our proposed EvoSkillRec.

cs.IR↗

arXiv2026

Successive Refinement Under Strong-Sense Perfect Perception

We revisit a multiterminal lossy source coding problem named successive refinement and derive the rate-distortion-perception region under the strong-sense perfect perception constraint in the presence of unlimited common randomness. Specifically, in successive refinement, one aims to compress a source sequence and allows two distinct decoders to recover the source sequence at different distortion levels. By imposing the strong-sense perfect perception constraint, our results refine the previous result by analyzing the impact of the perceptual quality. Our achievability proof is inspired by output constrained lossy source coding and our converse proof adapts the proof steps of the standard successive refinement problem. Furthermore, we provide a numerical example of the Bernoulli source to illustrate our result and show that the Bernoulli source under Hamming distortion is successively refinable even with the strong-sense perfect perception constraint.

cs.IT↗