arXiv subjects

Ziyu Liu

Publications and source records attributed to Ziyu Liu.

At least 19 recordsLinked to original sources

arXiv2026

Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models

Evaluating human-AI co-design requires testing whether coding agents can translate real users' multimodal requests into working application revisions. We introduce Co-Design Bench, a reference-free benchmark of 273 tasks curated from 37,463 deduplicated queries across 26,314 real user coding sessions. Its two tracks, Multimodal Editing (ME) and Repository Editing (RE), cover localized edits with optional image, audio, video, or document inputs and coordinated repository revisions. Tasks span three capability dimensions: Grounded Understanding, UI Transformation, and Functional Implementation. The tasks pair normalized requests and runnable pre-edit applications with 2,331 hidden expert-authored acceptance criteria. Our judge-then-refute protocol grades criteria using source and browser evidence, challenges positive verdicts with a separate LLM verifier, and audits visual and interaction claims without gold code or target screenshots. Under a shared agentic harness, the strongest of ten model endpoints passes an average of 69.9% of criteria per task but fully satisfies only 26.4% of tasks. All ten endpoints achieve lower criterion completion on repository edits spanning three or four change types than on those spanning one or two. Co-Design Bench provides a testbed for studying the gap between partial requirement satisfaction and complete task success under agentic execution for real-user application revisions.

cs.AI↗

arXiv2026

InstantFusion: A Shared Latent Interface for Heterogeneous Image Generators

Diffusion/flow models have emerged as the de facto paradigms for image generation, giving rise to a diverse family of modern generators with increasingly heterogeneous architectures and representations. Despite their remarkable generation capabilities, the cross-model alignment of latent representations and the consistency of their underlying denoising dynamics remain underexplored. In this work, we uncover an intriguing cross-model compatibility: independently trained generators exhibit similar latent spatial organization, while directly handing off intermediate states between unaligned models can still preserve coarse visual content. Motivated by these observations, we propose InstantFusion, a lightweight shared latent interface that connects the denoising trajectories of heterogeneous image generators. Specifically, to enable flexible cross-model state translation, model-specific Latent AutoEncoders (LAEs) are introduced to project intermediate states from different generators into a shared latent space and decode them into the latent space of a target generator. This shared space is constrained by two complementary objectives: aligning intermediate states derived from the same image at matched noise levels and synchronizing their corresponding local denoising velocities. Once trained, InstantFusion allows denoising to seamlessly transition across frozen generators without requiring a dedicated mapping for every model pair. The versatility of InstantFusion is demonstrated across acceleration, multi-preference composition, cross-architecture on-policy distillation, and cross-model composition, achieving competitive performance across these diverse settings.

cs.CV↗

arXiv2026

ResOPD: Tail Residualization for Sparse On-Policy Distillation

On-policy distillation (OPD) trains a student model on self-generated trajectories, but transmitting dense teacher distributions across long reasoning traces creates prohibitive communication and memory bottlenecks. Practical systems therefore rely on sparse teacher interfaces, typically transmitting either the sampled-token score or a small Top-$k$ distribution. However, this sparse setting faces a fundamental dilemma: sampled-token estimators are unbiased but suffer from severe gradient variance, whereas directly optimizing Top-$k$ objectives introduces systematic bias. To improve this trade-off, we propose ResOPD (On-Policy Distillation with Tail Residualization), which provides unbiased full-vocabulary reverse KL gradient estimation under on-policy sampling, with substantial variance reduction under the same sparse payload in the evaluated settings. ResOPD aggregates the unobserved vocabulary into an observable coarse tail event, computes its exact aggregate gradient, and samples only the fine-grained within-tail residual, which requires no additional teacher queries or forward passes. Extensive experiments demonstrate that ResOPD substantially reduces gradient variance, stabilizes online training dynamics, and improves downstream performance across the evaluated settings. These results establish ResOPD as an efficient, plug-and-play variance reduction primitive for sparse on-policy distillation.

cs.LG↗

arXiv2026

Exploiting Vulnerabilities: Universal Adversarial Attacks on Vision-Language-Action Models in Robotics

Recently, Vision-Language-Action (VLA) models have revolutionized robotic manipulation by seamlessly integrating visual perception, language understanding, and action generation in an end-to-end learning framework. However, since these models are designed to interact directly with the physical world and humans, their security is critical, and even small vulnerabilities can lead to catastrophic failures. In this work, we propose the Universal Adversarial Object, a sphere with optimized surface texture that significantly degrades task success rates when placed within the robot's field of view. Specifically, our approach introduces a multi-level attack framework that jointly disrupts trajectory planning, task execution, and action control. We validate our method in both simulated and real-world robotic settings. Experimental results demonstrate that the adversarial object reduces the average task success rates by 31.2%-39.9% for two representative VLA models (Pi0 and RDT), with success rates dropping to near zero in complex scenarios. Index Terms--Vision-Language-Action models, adversarial attack, robotic security, universal adversarial object

cs.RO↗

arXiv2026

Semantic Projection for Continual Self-Evolution of Language Agents

Language-model agents increasingly rely on persistent natural-language skills to adapt beyond their frozen model parameters. When a shared skill is repeatedly revised from a non-stationary, heterogeneous task stream, however, improvements for new tasks can overwrite procedures needed for earlier ones. In continual learning, Orthogonal Gradient Descent (OGD) addresses analogous interference by projecting a new-task gradient onto a subspace that locally preserves prior predictions. Natural-language skill revisions, however, have neither gradients nor a canonical vector space in which such a projection can be performed. We introduce \emph{Semantic-Scope Projected Evolution} (SSPE), which transfers the functional principle of gradient projection from parameter space to behavior space. SSPE treats an unconstrained skill revision as a proposed update, identifies acquired capabilities with which it may interfere, and uses the observed gains and regressions to construct a compatible revision rather than merely rejecting the update. This enables one shared skill to evolve across latent and recurring task contexts without exposing semantic domain identities to the evolution model. Across controlled synthetic streams and heterogeneous real-agent benchmarks, SSPE improves final cross-domain competence and mitigates forgetting relative to strong skill-evolution baselines. The evolved skill also retains the strongest average performance after transfer to a different executor model. These results establish semantic projection as a promising principle for stable and adaptive self evolution of language agents.

cs.AI↗

arXiv2026

Long-range self-avoiding walk in one dimension: a Monte Carlo study

We study the one-dimensional long-range self-avoiding walk in the grand-canonical ensemble where the statistical weight of a jump of length $r$ decays algebraically as $r^{-(d+σ)}$. Using large-scale Monte Carlo simulations with an efficient irreversible update scheme, we obtain high-precision estimates of the critical fugacity $z_c$, the universal Binder ratio $Q_N^c$, the correlation-length exponent $ν$, and the anomalous dimension $η$. For $σ>1$, the critical fugacity varies smoothly with $σ$, while the Binder ratio and the critical exponents remain consistent with the short-range universality class. For $σ\le 1$, by contrast, the results clearly depart from short-range behavior, identifying $σ=1$ as the boundary between long-range and short-range regimes. In the long-range Wilson-Fisher regime with $1/2 < σ\leq 1$, $η$ agrees with the long-range Gaussian-fixed-point prediction $η_{\mathrm{GFP}}=2-σ$, whereas $Q_N^c$ and $ν$ vary nontrivially with $σ$ and exhibit discontinuous jumps at $σ=1$. These findings are in good agreement with the recently proposed universality diagram in the $(d,σ)$ plane for long-range O$(n)$ models, with the self-avoiding walk corresponding to the $n\to0$ limit.

cond-mat.stat-mech↗

arXiv2026

Complex Problem Solving in Large Language Models: A Statistical Control Survey and Diagnostic Framework

Complex problem solving (CPS) with large language models (LLMs) is often framed as a matter of stronger reasoning or longer generation. Yet early-step error amplification, prompt brittleness, and failures to revise incorrect commitments are difficult to explain by missing knowledge or expressive capacity alone. This survey interprets CPS as a sequential estimation-and-decision problem over a latent solution state. A controller maintains a belief about an unobserved solution trajectory, updates it as noisy intermediate evidence arrives, and decides whether to commit, verify, branch, roll back, or abstain to minimize expected loss. Reasoning supplies candidate transitions and interpretations, whereas process control shapes and evaluates those proposals and regulates subsequent transitions and observations. Within this framework, we organize existing methods around five components: explicit state representation, transition structuring, validation and constraint enforcement, search and rollback, and uncertainty management. We also interpret evaluation metrics according to the statistical quantities they estimate. The framework further yields a diagnostic hypothesis: interventions should be most effective when they target the error or uncertainty component implicated by an observed failure. We distinguish systematic, stochastic, and irreducible error together with epistemic and aleatoric uncertainty, and call this alignment problem-control fit and its failure control mismatch. For example, additional sampling may reduce sampling variability while leaving a shared systematic error unchanged. This perspective clarifies what current methods estimate and control, what remains uncontrolled, and why reliable validation, targeted recovery, calibrated uncertainty, and matched-budget evaluation are central open problems.

stat.ML↗

arXiv2026

MAETrack: Unleashing the Potential of Pretrained Geometric Priors for 3D Single Object Tracking

Large-scale pre-training has transformed representation learning in 2D vision, yet its transferability to 3D single object tracking (SOT) remains insufficiently understood. Directly fine-tuning self-supervised 3D encoders, such as masked autoencoders (MAE), often leads to sub-optimal adaptation because the reconstruction objective is not fully aligned with the spatial-temporal matching requirements of tracking. In this paper, we observe that this difficulty can be interpreted as a layer-wise transfer mismatch: shallow layers tend to preserve transferable geometric cues, while deeper layers become increasingly specialized to the reconstruction pretext task and are less suitable for downstream tracking. Based on this observation, we propose MAETrack, a lightweight adaptation framework for transferring pre-training MAE representations to 3D SOT. MAETrack includes Layer-Selective Initialization (LSI), which initializes only the shallow stages of the tracking backbone from pre-trained weights while re-initializing deeper stages, and Geometric Residual Gating (GRG), which reinforces structurally salient regions in the search BEV features before template-search fusion through residual spatial modulation. Extensive experiments on standard 3D SOT benchmarks show that MAETrack consistently improves upon vanilla fine-tuning baselines with limited computational overhead. More broadly, our results suggest that effective transfer from 3D reconstruction pre-training to 3D tracking is not merely a matter of partial fine-tuning, but depends on a tracking-oriented transfer principle that preserves shallow geometry while adapting deeper representations to the downstream objective.

cs.CV↗

arXiv2026

ICI-VLA: In-Context Imitation with Spatiotemporally Aligned Demonstrations for Vision-Language-Action Models

Vision-Language-Action (VLA) policies are commonly adapted to new manipulation settings through additional gradient updates, which limits rapid deployment when task-specific data or compute is scarce. We present ICI-VLA, a training and retrieval framework that equips a text-action VLM with few-shot test-time adaptation through in-context demonstrations. Unlike mainstream VLA designs based on action-specific multimodal fusion, ICI-VLA retains the native text-generation interface. ICI-VLA updates its parameters only during offline training; at inference, the policy remains fixed and conditions action generation on retrieved micro-demonstrations. The framework decomposes long trajectories into short, semantically labeled examples and trains an RD-Encoder with positives mined by Dynamic Time Warping (DTW), aligning the retrieved context with the phase and geometry of the current subtask. We further introduce Target Action Masking, a context-corruption objective designed to reduce direct action copying and increase reliance on the current observation. ICI-VLA reaches average success rates of 97.7% on LIBERO and 60.4% on RoboTwin 2.0, exceeding the highest reported baseline average on RoboTwin 2.0 by 19.3 percentage points. It also achieves 83.2% across four physical tasks. These results indicate that a fixed VLA policy can benefit from conditioning on spatiotemporally aligned demonstrations at test time.

cs.RO↗

arXiv2026

Antiferromagnetism-altered plasmon dynamics

The interaction between plasmons and magnons is a long-sought phenomenon with implications for fundamental physics and spintronics applications. In three-dimensional systems, this coupling is suppressed by the large mismatch in energy scales, but two-dimensional (2D) plasmons with gapless dispersion can overlap with magnons over a broad spectral range. Despite numerous theoretical predictions, experimental observation of magnon-plasmon interaction has remained elusive. In this work, we study a first-of-its-kind hybrid plasmon-magnon platform based on 2D materials. By deploying scattering-type scanning near-field optical microscopy (s-SNOM) with terahertz radiation, we image propagating plasmon wavepackets at a graphene/NiPS$_3$ interface and track their dynamics across the antiferromagnetic transition of NiPS$_3$. We observe a clear renormalization of the plasmon-polariton dispersion concurrent with the onset of antiferromagnetic order. With complementary Raman scattering and nano-terahertz spectroscopy, we unveil spectral weight redistribution and dielectric screening changes, potentially associated with the multi-magnon continuum, as the underlying mechanism. These results provide solid evidence of coupling between plasmon and antiferromagnetic order, marking a cornerstone for a potential platform for hybrid magnon-plasmon interactions in 2D materials, opening avenues for coherent spin-plasmon devices and tunable terahertz spintronic components.

cond-mat.str-el↗

arXiv2026

Exponential mixing for the stochastic Navier--Stokes equation with localized noise

This paper studies the 2D Navier--Stokes equation on a bounded domain with Navier-slip boundary conditions, driven by spatially localized stationary forcing generated by a stochastic heat equation. We prove exponential mixing for the associated Navier--Stokes--heat system, and consequently exponential convergence of the velocity law under stationary forcing. The main difficulty is that the white noise is confined to a subdomain and reaches the velocity indirectly through the heat component. To address the degeneracy, the proof combines Malliavin calculus with PDE control theory. The key ingredient is a stabilization scheme that converts localized Navier--Stokes controls into time-regular controls compatible with the heat dynamics.

math.PR↗

arXiv2026

When Emotion Becomes Trigger: Emotion-style dynamic Backdoor Attack Parasitising Large Language Models

Data-poisoning backdoors pose a practical threat to the fine-tuning of large language models (LLMs). Most existing attacks bind an attacker-selected behavior to fixed tokens, phrases, scenarios, or syntactic structures. These discrete triggers provide concrete handles for defenses based on local token anomalies, pattern matching, or trigger recovery. We found that, \emph{under semantics-preserving rewriting, emotionally styled inputs form representation clusters distinct from their neutral counterparts}. Meanwhile, de-emotionalised controls move back towards the neutral distribution. This observation motivates our method \textbf{Paraesthesia}, a dynamic backdoor attack that encodes its triggering condition in an emotional style. Paraesthesia maps target emotions into a valence--arousal space, rewrites a small subset of clean samples, and retains semantically faithful rewrites for fine-tuning. Across instruction-following and classification tasks evaluated on four major LLMs, Paraesthesia achieves an attack success rate(ASR) above 98.25\%, while introducing only negligible degradation to clean utility across the vast majority of model-task setups. Surface feature controls and paired de-emotionalization experiments demonstrate that no examined token-level cue can fully account for the triggered behavior. ASR remains high after word-level filtering, sample clustering, and subsequent clean-update procedures, whereas a white-box decoding defense with access to a task-aligned clean reference provides a distinct mitigation path. These findings identify emotional style as a concrete backdoor trigger surface beyond fixed lexical and syntactic patterns.\par\smallskip \noindent \textcolor{red}{\textbf{WARNING: }\textnormal{This paper contains risk-related textual content.}}

cs.CL↗

arXiv2026

Two-dimensional percolation with algebraically decaying interactions II: Critical exponents in the long-range regime

We present a comprehensive Monte Carlo study of two-dimensional bond percolation with algebraically decaying connection probabilities $p(r)\propto 1/r^{2+σ}$, establishing the universality diagram in the long-range (LR) regime for $σ\le2$. Using the event-based ensemble method, we simulate systems with linear sizes up to $L=16384$ and investigate three universality regimes: LR Wilson--Fisher (WF) A ($1<σ\le2$), LR Wilson--Fisher B ($2/3<σ\le1$), and LR mean-field (MF) ($0<σ\le2/3$). In the LR-WF-B regime, the anomalous dimension is consistent with $η=2-σ$, in agreement with mathematical results for $2/3<σ<1$, while the correlation-length exponent $ν(σ)$ exhibits nontrivial, non-Gaussian variation. In the LR-WF-A regime, although $η$ remains close to $2-σ$ for smaller $σ$, statistically resolvable deviations $δη(σ)=η-(2-σ)>0$ start to appear near $σ\simeq3/2$ and grow toward the short-range crossover at $σ=2$. Finally, by complementing the event-based simulations with conventional ensemble simulations, we reveal the coexistence of complete-graph asymptotics and LR Gaussian-fixed-point scaling in the LR-MF regime. These results further clarify the critical properties in long-range percolation and provide crucial benchmarks for long-range statistical systems.

cond-mat.stat-mech↗

arXiv2026

Intern-S2-Preview: Scientific Agentic Foundation Model

Scientific discovery increasingly requires AI systems that can reason over scientific evidence of heterogeneous modalities, interact with scientific tools and environments, and sustain progress across long task horizons. We present Intern-S2-Preview, a series of scientific agentic foundation models designed to support multimodal scientific understanding, reasoning, generation, and long-horizon tasks. The training pipeline begins with scientific multimodal pre-training over rendered scientific documents, interleaved image-text data, and diverse scientific corpora. Starting from the pretrained checkpoint, we apply a unified post-training pipeline consisting of supervised fine-tuning, scalable multi-task reinforcement learning (RL), black- and white-box agentic RL, and on-policy distillation. This pipeline is supported by practical techniques that improve rollout and training stability and efficiency, including partial rollout with off-policy correction, adaptive length regularization, online speculative decoding, robust multi-task optimization, and trace-aware experience assembly for agentic tasks. At the architecture level, Intern-S2-Preview-397B extends time series modelling from efficient long-sequence understanding to numerical forecasting, while Memory Decoder is studied as a separate memory-augmented path for rapid scientific specialization without modifying the frozen 397B backbone. Evaluations across scientific, multimodal, agentic, and general-purpose benchmarks show that Intern-S2-Preview-397B achieves competitive or leading results in multiple settings. The time series modules improve scientific signal understanding and forecasting on SciTS, while the separate Intern-MemDec-4B extension improves the Biology-Instructions average score from 56.92 to 60.32 without modifying the frozen 397B backbone.

cs.LG↗

arXiv2026

RIVERPlace: Repairing Interconnect Violations with Efficient Retiming and Incremental Placement for AQFP Circuits

The Adiabatic Quantum-Flux-Parametron (AQFP) offers near-Landauer-limit energy efficiency but faces significant scalability challenges due to strict path balancing and limited drive strength. To address this, we propose RIVERPlace, a framework that integrates long-wire pipelining, retiming, and incremental placement to resolve interconnect violations with minimal disruption. RIVERPlace first applies placement-aware retiming to repair violations without increasing logical depth. When depth increases are necessary, we introduce Buffer Cut Insertion (BCI), which formulates violation resolution as a constrained global edge-selection problem reducible to a maximum topological cut, thereby enabling an exact polynomial-time solution. By selectively pipelining edges across multiple rows, BCI avoids excessive buffer insertion while resolving interconnect violations. Experimental results demonstrate that RIVERPlace consistently outperforms prior AQFP placement approaches, reducing placement overhead by more than an order of magnitude in inserted buffers, 3x in placement-induced depth, and over 2x in circuit area, while also reducing runtime by more than an order of magnitude and latency by 38%. These improvements enable the first post-routing, timing-closed implementations of the complete open-source AQFP benchmark suite, including larger circuits like alu32.

cs.ET↗

arXiv2026

On the classification of invariant Gaussian measures for the 2D Euler equation

We consider the two-dimensional incompressible Euler equation on $\mathbb T^2$ with Gaussian random initial data having independent Fourier coefficients. For every $σ>0$, we prove that such a Gaussian measure on $H^σ(\mathbb T^2)$ is invariant under the Euler flow if and only if it is supported either on shear flows or on cellular flows. This settles the invariant-measure classification conjecture posed by Bedrossian and Latocca (Ann. Inst. H. Poincaré C Anal. Non Linéaire, 2026). The proof relies on closure of the Fourier support under non-degenerate interactions and an affine relation among the inverse variances along Euler triples.

math.AP↗

arXiv2026

Bridging AQFP Technology Legalization and Physical Design: Layout-Aware Buffer and Splitter Insertion via Width--Depth Product Minimization

Adiabatic Quantum Flux Parametron (AQFP) is an emerging superconducting technology that enables ultra-low energy dissipation approaching the Shannon limit. However, its gate-level pipelining and explicit fanout constraints require technology legalization through buffer and splitter insertion to ensure path balancing and signal distribution, becoming a critical and costly step in the design flow. Prior work has focused on minimizing inserted cell count and logic depth, yet these objectives do not accurately capture the final physical design cost, which is fundamentally governed by the product of circuit width and depth. In this article, we redefine AQFP buffer and splitter insertion optimization as minimizing the circuit width--depth product, a layout-aware metric that more accurately captures physical design area than prior cell minimization efforts. We are the first to formulate buffer and splitter insertion under this objective and prove that the resulting problem is NP-complete. To address this complexity, we develop scalable heuristics that integrates legalization with this objective. Experimental results on standard benchmarks demonstrate that our approach achieves an average 30% reduction in post-placement area compared to state-of-the-art methods, with only a 3% increase in junction count, and on individual circuits up to 61% area reduction, demonstrating the effectiveness of the proposed objective in reducing true design cost.

cs.ET↗

arXiv2026

Asymptotic strong Feller and weak observability inequality

For a class of non-autonomous linear SPDEs, we establish the equivalence among asymptotic regularization, weak observability, and approximate null controllability for the associated deterministic control systems. This equivalence provides a deterministic control-theoretic characterization of stochastic smoothing and offers a systematic approach to studying SPDEs driven by spatially localized noise. We further establish a criterion for semilinear SPDEs based on weak observability of the linearized equations. Our approach combines methods from PDE control theory with Malliavin calculus. As applications, we consider the stochastic Oseen equation, non-autonomous uniformly parabolic equations, and the parabolic Sine--Gordon equation, all driven by finite-dimensional, spatially localized white-in-time noise.

math.PR↗