Searcharxiv⌕ Search

arXiv subjects

Yujie Zhang

Publications and source records attributed to Yujie Zhang.

At least 19 recordsLinked to original sources
arXiv2026

3DGS-VBench+ and 3DGSV-Assessor: A Multi-Dimensional Subjective Benchmark and a Reference-Flexible Objective Model for 3D Gaussian Splatting Video Quality Assessment

3D Gaussian Splatting (3DGS) enables real-time novel view synthesis (NVS), but its heavy storage makes compression essential, and both training and compression introduce representation-specific distortions. Since generative compression in many methods treats geometric and color attributes via different parameters, it can induce dimension-specific distortions, motivating multi-dimensional assessment. Moreover, real-world captures generally lack an aligned ground-truth video for a synthesized, continuous camera trajectory, leaving only sparse captured views as references. To enable multi-dimensional quality assessment of 3DGS videos with scarce aligned references, we first present 3DGS-VBench+, a large-scale subjective video quality assessment (VQA) benchmark comprising 3{,}036 3DGS reconstructions, each paired with a rendered video, across 11 real-world and 22 synthetic scenes, generated by 6 representative compression algorithms at systematically designed levels and annotated with 9{,}108 MOSs covering overall, geometry, and color quality. Based on 3DGS-VBench+, we propose 3DGSV-Assessor, a reference-flexible, large multimodal framework that scores all three dimensions in one generation pass and fuses a ranking-based reference cue into the frozen regressor through a bounded gate, supporting both reduced-reference (RR) and no-reference (NR) assessment. Extensive experiments demonstrate state-of-the-art correlation with human perception on 3DGS-VBench+ and strong zero-shot generalization to other NVS benchmarks. The dataset and code will be released at https://github.com/YukeXing/3DGSV-Assessor.

cs.CV↗
arXiv2026

MedVTok: A General-Purpose Medical Visual Tokenizer

Multimodal medical AI requires a common visual interface for encoding heterogeneous imaging evidence for patient condition modeling. However, existing medical visual tokenizers are often tied to a specific dimension and imaging modality, forcing multimodal systems to rely on fragmented and poorly reusable representations. We introduce MedVTok, a general-purpose visual tokenizer that maps 2D images and 3D volumes from diverse imaging modalities into a unified token space. Building such a tokenizer poses two key challenges: preserving anatomical structure across dimensions and capturing clinical semantics across modalities. To this end, MedVTok introduces (i) slice differential regularization, which explicitly models inter-slice anatomical coherence missing in previous per-pixel/per-voxel supervision, and (ii) multi-expert representation alignment, which integrates knowledge from multiple medical expert encoders while retaining modality-specific complementary knowledge. Trained on 28M images and 70K volumes across 9 imaging modalities, MedVTok supports classification, retrieval, segmentation, synthesis, visual question answering, and medical report generation through a single tokenizer. Across 24 benchmarks, MedVTok achieves state-of-the-art performance, demonstrating a scalable token interface that unifies medical image dimension, modality, and task usage.

eess.IV↗
arXiv2026

EpiPersona: Persona Projection and Episode Coupling for Pluralistic Preference Modeling

Pluralistic alignment is essential for adapting large language models (LLMs) to the diverse preferences of individuals and minority groups. However, existing personalized reward models often mix stable personal traits with episode-specific factors, limiting their ability to generalize across episodes. To address this challenge, we propose EpiPersona, a framework for explicit persona-episode coupling. EpiPersona first projects noisy preference feedback into a low-dimensional persona space, where similar personas are aggregated into shared discrete codewords. This process separates enduring personal characteristics from situational signals without relying on predefined preference dimensions. The inferred persona representation is then coupled with the current episode, enabling episode-aware preference prediction. Extensive experiments show that EpiPersona outperforms the baselines. It achieves notable performance gains in hard episodic-shift scenarios, while remaining effective with sparse preference data.

cs.AI↗
arXiv2026

Move First, Commit Later: Selective LiDAR-to-BIM Global Initialization by Verifying Proposals Against Competing Explanations

Global LiDAR-to-BIM initialization must place a robot within an as-designed building model without a prior pose. In repetitive interiors, the principal failure mode is not low-confidence registration but confident aliasing: a submap can match several BIM regions with comparable scores, producing a high-scoring pose displaced by symmetry. We present Move First, Commit Later, a selective layer that treats a registration front-end as an evidence source and decides whether to commit. Candidates from multiple submaps are mapped to a common $\mathrm{SE}(2)$ anchor; a top-1 consensus $B_m$, invariant to non-champion multiplicity, aggregates cross-submap evidence; and topology serves only as a binary feasibility gate. The decision is typed--COMMIT, DEFER, or AMBIGUOUS($τ$), reporting the detected symmetry period--and reversible: symmetry-breaking motion upgrades AMBIGUOUS to COMMIT. On a real multi-room building and a controlled symmetric simulation, the layer commits correctly in every tested trajectory-scale configuration and otherwise abstains with a typed state, whereas forced-choice policies on the same front-end select wrong rooms in most cases. Committed anchors are within 0.02--0.36 m of an independent laser-tracker position reference and within $1.9^\circ$ of a BIM-registration orientation proxy. The evaluation covers one building and one front-end; the layer is designed to be front-end modular.

cs.RO↗
arXiv2026

Neos: Learning to Interleave Search and Visual Reasoning

Recent multimodal deep search agents have advanced tool use and visual content retrieval, yet they still struggle to convert intermediate visual evidence into actionable guidance for subsequent search, limiting their effectiveness in long-horizon settings. To address this gap, we present Neos, a multimodal deep search framework that learns both how to acquire evidence and how to use it to guide subsequent decisions. To make this coupling learnable, Neos constructs training tasks with intermediate visual dependencies while filtering out textual shortcuts, and introduces a lightweight, trainable Multimodal Evidence Interface (MEI) that transforms retrieved content into goal-conditioned observations. At its core, an agent-environment co-evolution procedure treats MEI as a co-adaptive component of this loop. Specifically, reference-guided credit assignment uses verified successful trajectories to localize critical policy errors while preserving credit for useful decisions in failed searches, whereas rollout feedback selects informative local evidence-processing interactions for MEI supervised fine-tuning (SFT). By alternating policy reinforcement learning (RL) with MEI SFT, Neos allows the policy's evolving evidence requests to shape MEI training and the adapted MEI to reshape subsequent observations, closing the loop between evidence acquisition and evidence-conditioned action. Across ten benchmarks, Neos-9B and Neos-35B demonstrate leading performance among specialized multimodal deep-search agents.

cs.AI↗
arXiv2026

Experience Makes Skillful: Enabling Generalizable Medical Agent Reasoning via Self-Evolving Skill Memory

Medical agent systems are increasingly expected to support interactive clinical decision making rather than only static question answering. In such settings, effective agents must reuse prior experience across evolving cases, yet existing memory mechanisms often retain raw historical traces that are redundant, noisy, and difficult to govern. More importantly, they rarely distinguish which memories are truly useful for future reasoning. This limits their ability to accumulate compact and reliable experience for long-horizon clinical reasoning. To close this gap, we propose SkeMex, a post-deployment self-evolution framework that improves medical agents through a skill-based memory without updating model weights. SkeMex distills informative interaction trajectories into structured skills that encode reusable procedural knowledge, and organizes them into a multi-branch repository spanning general, task-specific, and action-level experience. To determine which memories should be reused and retained, SkeMex estimates context-dependent utility from environment feedback and uses it to guide value-aware retrieval and repository governance. A closed-loop ``Read--Write--Assess--Govern" lifecycle further supports continual evolution by writing new skills, updating utilities, promoting useful memories, and removing harmful entries. Experiments across diverse clinical tasks show that SkeMex consistently outperforms representative memory-based agents in both offline and online settings. It also generalizes across model backbones and supports transferable skill memory. All data and code will be released publicly.

cs.AI↗
arXiv2026

Complexity of self-consistent entanglement certification

In recent work Phys. Rev. X 16, 031057 (2026), we proposed a self-consistent approach to entanglement certification based on generalized noncontextuality. It requires no prior characterization of the measurement devices and, given access to all local measurements, can certify every entangled state in a Bell circuit. Here, we study the complexity of this protocol in a finite experiment. At fixed local dimension $d$, we show that the optimal number of distinct local effects needed to reach a trace-distance \textit{entanglement resolution} $γ$ is $Θ_d(γ^{-(d-1)})$. This scaling is necessary even when the measurements are tailored to the target state. Independent Haar-random projective measurements instead require $Θ_d(γ^{-(d-1)}\log(1/γ))$ effects. Finally, we show that, once the exact operational identities are known, $Θ_d(γ^{-2}\log(1/δ))$ copies of the target state are necessary and sufficient for certification with error probability at most $δ$. Determining the optimal sample complexity when those identities must instead be inferred from the same finite target-state data remains open.

quant-ph↗
arXiv2026

Destroying and Preserving Measurement Incompatibility over a Quantum Channel

A quantum channel is called incompatibility-breaking if it always destroys the incompatibility of quantum measurements. On the other extreme are incompatibility-preserving channels, which never destroy incompatibility. This work studies both types of channels through the mathematical framework of zonotopes and zonoids, an approach that provides a complete characterization of compatibility for dichotomic measurements and is particularly powerful in the qubit setting. Using the language of zonoids, we present necessary and sufficient conditions for when a qubit unital channel is incompatibility-breaking, which are also sufficient for nonunital channels. As one of our main results, we demonstrate the superactivation phenomenon of two incompatibility-breaking qubit channels that are no longer incompatibility-breaking when used in parallel. Translating this result into the task of quantum steering, it says that two copies of a two-qubit Werner state can be steered at noise threshold $1/2$, even though this is impossible using just one copy. For the problem of incompatibility preservation, we prove that a quantum channel with matching input and output dimensions will preserve general incompatibility if and only if it is a unitary transformation. When the output dimension is larger, this result no longer holds, and we offer a conjectured form of a general incompatibility-preserving channel.

quant-ph↗
arXiv2026

Compatibility of quantum instruments

Measurement compatibility asks whether a family of quantum measurements can be simulated by one parent measurement followed by classical post-processing. We extend this measurement compatibility to the notion of \emph{$d$-compatibility} for measurements, channels, and instruments whose parent passes a $d$-dimensional quantum system and unlimited classical information to the post-processing. We formulate $d$-compatibility as a common linear factorization of operator spaces and operator systems and show that $d$-compatible families form a closed set, even with infinitely many settings and a separable infinite-dimensional input. To certify $d$-incompatibility, we define the compatibility functional as the smallest product of completely bounded norms over common factorizations through a $d$-dimensional register. Using factorization theory of operator spaces, we show that measure-and-prepare channels defined in terms of mutually unbiased bases and pinching channels defined in terms of Heisenberg-Weyl operators are $d$-incompatible in certain regimes of $d$. We also link $d$-compatibility of multi-instruments to the preparability of Choi assemblages using channel-state duality, and use this connection to show that for bipartite states the (in)compatibility of instruments can be used to certify Schmidt numbers and the implementability of the instrument using one-way LOCC protocols. Finally, our compatibility functional also bounds the generalized robustness of compatibility, which we interpret through a memory-bounded nontransient preparation game.

quant-ph↗
arXiv2026

OuroReward: Sequential Reward Scheduling for Reinforcement Learning in Text-to-3D Generation

Reinforcement learning (RL) for Text-to-3D (T23D) generation requires optimization across multiple quality dimensions such as semantic alignment and texture clarity. Existing methods typically optimize these dimensions simultaneously through multiple reward aggregation, without explicitly modeling inter-dimension dependencies. This can cause imbalanced optimization and persistent interference among conflicting dimensions. To address this limitation, we propose OuroReward, an interference-aware sequential reward scheduling strategy for T23D RL. OuroReward first estimates pairwise dependencies among dimensions and constructs a cyclic optimization path that minimizes cumulative interference. By incorporating the tail-to-head dependency, the cycle captures global compatibility across the entire schedule. Then, OuroReward converts the cycle into a one-pass sequence, and starts optimization from the dimension with the lowest aggregate interference. Rather than assigning a fixed optimization budget to each dimension-wise reward, training adaptively determines when to advance to the next reward according to the remaining optimization headroom of the current one. We further introduce AdaSelect, an adaptive prompt selection strategy that identifies reliable and informative prompts aligned with the model's current capability. By focusing policy updates on these prompts, AdaSelect effectively improves training stability. Extensive experiments across different T23D models and RL algorithms demonstrate that our framework consistently improves generation quality across multiple dimensions.

cs.CV↗
arXiv2026

CoEvolve: Construct-to-Edit Visual Grounding with Bidirectional State Refinement

Visual grounding localizes an object described by language with a bounding box. Most multimodal grounding models compress target identification, spatial reasoning, and boundary estimation into one terminal prediction. Free-form rationales make reasoning linguistically explicit but do not necessarily expose measurable, editable spatial states. Intermediate localization errors are therefore difficult to diagnose and correct, allowing incorrect region choices and imprecise boundaries to persist in the final box. We introduce CoEvolve, a construct-to-edit framework that separates grounding into explicit state construction and state editing. Region-Evolution Reinforcement (RER) organizes grounding analysis into a progressive semantic--spatial trajectory, with each reasoning step committing to an explicit candidate region. Bidirectional Denoising Refiner (BDR) treats the reasoning text as fixed semantic context and refines the trajectory's coordinate fields through bidirectional same-position reconstruction. Geometry- and behavior-level objectives provide target geometry and edit-preference signals for consolidating reliable candidates, preserving accurate inputs, or correcting toward annotations. Evaluations cover natural-image and remote-sensing grounding. With a 9B backbone, CoEvolve rivals models up to 241B parameters in grounding accuracy. Under controlled corruption, a single BDR pass improves mean box overlap by over 27 percentage points, demonstrating strong recovery from substantial localization errors. State-source comparisons further support the complementarity of explicit state construction and source-matched editing. The project is at https://sundongwei.github.io/CoEvolve_Project/.

cs.AI↗
arXiv2026

ACE: Agentic Control for Embodied Manipulation via Zero-shot Workflow Reasoning

General-purpose manipulation requires both semantic reasoning over task constraints and reliable execution of contact-rich actions. We present ACE, an agentic manipulation harness that composes a high-level language agent with a reusable mask-conditioned visuomotor policy. Given an open-ended instruction, the agent solves semantic constraints, binds objects to destination roles, and decomposes the task into executable transfers represented by tracked pick-and-place masks. Execution feedback supports outcome assessment, re-grounding, and retry, while persistent object and task context preserves earlier associations when manipulation changes visible cues. We evaluate ACE on two physical multi-step tabletop tasks, Semantic Formula Assembly and Constraint Retrieval. The visuomotor policy is trained only on generic pick-and-place demonstrations and reused without complete demonstrations of either evaluation task, enabling task-level zero-shot composition. Across 20 randomized trials per task, ACE achieves 70% and 80% success, respectively, compared with 55% and 70% without persistent context. These results suggest that an agentic harness can extend a primitive-trained manipulation policy to semantically distinct tasks through explicit object-destination interfaces and closed-loop execution feedback.

cs.RO↗
arXiv2026

XS-VLA: Teaching Tiny Vision-Language-Action Models with Spatial Supervision and Demonstration Conditioning

How can richer training supervision improve robot control while keeping the deployed policy compact? We present XS-VLA, a staged training framework using teacher-derived spatial labels and demonstration-conditioned action learning. Coarse-Grained Spatial Distillation (CSD) initializes the backbone through an auxiliary region-label task. Latent Flow Matching (LFM) then conditions an action-space velocity field on a demonstration latent, using KL regularization while jointly optimizing the backbone and action modules. The deployed policy contains 243.99M parameters and operates without the teacher or posterior encoder. XS-VLA achieves 90.25% average LIBERO success in each of two training seeds, compared with 86.00% for a SmolVLA-256M base trained under our settings. Ablations examine both training stages through matched image pretraining and Huber/MSE controls. On three Mobile ALOHA tasks, average strict success increases from 21.7% to 65.0%. These results demonstrate the control utility of auxiliary representation initialization and regularized demonstration-conditioned flow learning for compact VLA~policies.

cs.RO↗
arXiv2026

Quantum key distribution using generalized contextuality against post-quantum eavesdroppers

Bell nonlocality allows secret key to be certified without trusting the internal operation of the devices, and even against eavesdroppers constrained only by the no-signaling principle. We ask whether generalized contextuality, a notion of nonclassicality available in prepare-and-measure experiments on a single system, can play an analogous role. In our approach, Alice and Bob first establish operational equivalences among their local preparation and measurement procedures, which are then treated as theory-independent constraints that any physical intervention on the communication channel, including that of an adversary, must respect. Eve is otherwise unrestricted, that is, her intervention on the transmitted system is not assumed to be described by quantum theory. Within this adversarial model, we show that Eve's optimal guessing probability against individual attacks is the solution of a finite linear program, and hence that a worst-case asymptotic key rate follows from the observed statistics and the operational equivalences alone. We prove that contextuality is necessary for a positive key rate. However, we also show that it is not sufficient. Concretely, the (3,2)-parity-oblivious random access code, known to yield secret key against a quantum eavesdropper, yields none certifiable key rate against a post-quantum one. Conversely, operational equivalences with no analogue among no-signaling constraints can strictly increase the asymptotic key rate. We illustrate this with the prepare-and-measure version of the standard CHSH protocol with an aligned key-generating setting.

quant-ph↗
arXiv2026

Reassessing the boundary between classical and nonclassical for individual quantum processes

There is a received wisdom about where to draw the boundary between classical and nonclassical for various types of quantum processes. For multipartite states, it is the divide between separable and entangled; for channels, the divide between entanglement-breaking and not; for sets of measurements, the divide between compatible and incompatible; for assemblages, the divide between steerable and unsteerable. However, these choices have not been motivated by any unified notion of what it means to be classically explainable. One well-motivated notion of classical explainability is the one based on generalized noncontextuality: a set of circuits is classically explainable if a generalized-noncontextual ontological model can realize the statistics they generate. In this work, we show that this notion can be leveraged to define a classical-nonclassical divide for individual quantum processes of arbitrary type. We begin the task of characterizing where the classical-nonclassical divide lies according to this proposal for a variety of different types of processes. In particular, we show that all of the following are judged to be nonclassical: every entangled state, every set of incompatible measurements, every non-entanglement-breaking channel, and every steerable assemblage. Our proposal differs from the received wisdom, however, insofar as it also judges certain subsets of the complementary classes to be nonclassical, including certain separable states, compatible sets of measurements, entanglement-breaking channels, and unsteerable assemblages. Finally, we prove structure theorems characterizing the classical-nonclassical divide based on whether a process admits of a specific type of frame representation.

quant-ph↗
arXiv2026

Entanglement certification using noncontextuality inequalities

By combining the assumptions of Bell locality with those of generalized noncontextuality, we define classes of noncontextuality inequalities for correlations arising in a bipartite Bell circuit. These classes are distinguished by which subsets of the full set of operational identities are taken as input to the principle of noncontextuality; certain natural subsets form a hierarchy that provides a new way of understanding and classifying quantum correlations, including entanglement, steering, and nonlocality. Each level of this hierarchy gives rise to a corresponding class of noncontextuality inequalities whose violation witnesses one of these forms of bipartite quantum resourcefulness, thereby yielding different sufficient conditions for entanglement. The resulting entanglement certification paradigm requires no prior characterization of the measurements, is independent of tomographic gauge freedom, and can certify any entangled state without auxiliary entangled sources. To illustrate its power, we show that noncontextuality inequalities can certify entanglement for families of two-qubit isotropic states for which Bell or steering inequalities are known to fail. We also show that, compared with the Bell test, this approach certifies a much larger fraction of entangled states, while the associated membership problem is more tractable. On the experimental side, we describe techniques to ensure nontrivial operational identities in the presence of noisy and imperfect implementations. We also identify the key assumption under which these techniques are valid, namely, a particular notion of tomographic completeness, which ensures that the operational identities are gauge-independent. Finally, we provide an experimental demonstration of the superior performance of this entanglement certification technique using polarization-entangled photons.

quant-ph↗
arXiv2026

Para-Pipe: Exploiting Hierarchical Operator Parallelism of ML Computational Graphs on SoCs

As edge-based deep learning applications become more complex, optimizing performance on heterogeneous System-on-Chips (SoCs) presents unique challenges. Traditional pipelining techniques distributing the computation across different on-chip processing units, while effective for throughput, do not address the latency demands posed by modern neural networks with complex interdependencies and extensive operator parallelism. There is a potential in leveraging operator parallelism to enable concurrent execution across multiple processing units, thereby reducing inference latency. However, prioritizing pipelining or parallel execution often necessitates a compromise, where optimizing one performance metric adversely impacts the other. This paper introduces Para-Pipe, a hierarchical mapping framework that integrates intra- and inter-stage operator parallelism within a pipelined architecture. Para-Pipe navigates the trade-off between throughput and latency by selectively fine-tuning parallelism levels within and across pipeline stages. This strategy can significantly reduce inter-processor communication overhead, significantly improving energy efficiency. Our evaluation demonstrates that Para-Pipe generates multiple Pareto-optimal configurations, achieving a balance between throughput and latency on an Amlogic SoC equipped with ARM big.LITTLE CPUs and GPU, as well as the Black Sesame Technology SoC featuring a deep learning accelerator and two DSPs. More importantly, throughput-optimized configurations under Para-Pipe on Amlogic SoC show an average energy efficiency improvement of 11.0% over purely pipelined strategies and 23.3% relative to non-pipelined parallel execution.

cs.DC↗
arXiv2026

Towards Fine-Grained Text-to-3D Quality Assessment: A Benchmark and A Two-Stage Rank-Learning Metric

Recent advances in Text-to-3D (T23D) generative models have enabled the synthesis of diverse, high-fidelity 3D assets from textual prompts. However, existing challenges restrict the development of reliable T23D quality assessment (T23DQA). First, existing benchmarks are outdated, fragmented, and coarse-grained, making fine-grained metric training infeasible. Moreover, current objective metrics exhibit inherent design limitations, resulting in non-representative feature extraction and diminished metric robustness. To address these limitations, we introduce T23D-CompBench, a comprehensive benchmark for compositional T23D generation. We define five components with twelve sub-components for compositional prompts, which are used to generate 3,600 textured meshes from ten state-of-the-art generative models. A large-scale subjective experiment is conducted to collect 129,600 reliable human ratings across different perspectives. Based on T23D-CompBench, we further propose Rank2Score, an effective evaluator with two-stage training for T23DQA. Rank2Score enhances pairwise training via supervised contrastive regression and curriculum learning in the first stage, and subsequently refines predictions using mean opinion scores to achieve closer alignment with human judgments in the second stage. Extensive experiments and downstream applications demonstrate that Rank2Score consistently outperforms existing metrics across multiple dimensions and can additionally serve as a reward function to optimize generative models. The project is available at https://cbysjtu.github.io/Rank2Score/.

cs.CV↗