Search arXiv⌕ Search

arXiv subjects

Yuxuan Zhou

Publications and source records attributed to Yuxuan Zhou.

At least 19 recordsLinked to original sources
arXiv2026

When Does Better Acceptance Pay? A Controlled Study of Speculative Decoding

The throughput gains of speculative decoding depend on the trade-off between acceptance length and execution latency. We ask when additional acceptance is worth the cost of obtaining it. We establish a comparison protocol spanning training, inference, and evaluation, remeasuring released checkpoints and retraining six drafter recipes on shared target-generated data across ten benchmarks and two target models. Alignment changes the picture: gaps that appear large in published comparisons shrink substantially under shared training and inference conditions. This narrowing motivates two deeper questions: how is acceptance improved at different costs, and when do these gains improve throughput? We find that acceptance can be improved through nearly free objective and context changes or more expensive token-to-token conditioning, which pays off most clearly under sampled drafting. DCore illustrates the former route as a strong baseline with little additional measured drafting overhead. The payoff is conditional: task uncertainty and serving conditions affect acceptance returns, execution costs, and the preferred drafter and budget. These findings inform future drafter design and deployment choices.

cs.CL↗
arXiv2026

Dynamic Query as Budgeting for Tiny Object Detection

Recent DETR-based tiny object detectors adopt dynamic query mechanisms to handle severe density imbalance. However, existing designs entangle two effects: how many and which queries are decoded. This entanglement obscures what drives the gains and complicates dynamic-query design; additional learned predictors and heuristic filtering also introduce overhead and potential selection errors. We disentangle these effects and show that dynamic query primarily acts as \emph{budgeting}---allocating decoder query capacity according to object density---rather than intricate query selection.Motivated by this view, we formulate a monotonic and conservative budgeting principle and propose a budgeting-based DETR (BUTR), a streamlined framework that uses a lightweight, non-learned budgeter to estimate an input-dependent query budget $K(x)$ from encoder confidences. BUTR retains standard Top-$K(x)$ query initialization without additional learned modules or heuristic filtering in the budgeter.At $800\times800$ resolution, BUTR-L achieves 35.7 AP at 47.8 FPS on AI-TOD-v2 and 29.8 AP at 42.1 FPS on VisDrone on an NVIDIA T4 GPU. Inference-time integration on COCO further demonstrates its applicability to general scenario detection. Code is available at https://github.com/zhangyi-magic/BUTR.

cs.CV↗
arXiv2026

Sharp quantitative stability of the Ashbaugh--Benguria inequality

We establish sharp quantitative stability of the Ashbaugh--Benguria inequality on bounded convex domains in $\mathbb R^n$. After volume and centroid normalization, the eigenvalue-ratio deficit controls the squared $H^{-1/2}(\mathbb S^{n-1})$ norm of the radial displacement from the unit sphere. This yields the optimal exponent $5/2$ for the Fraenkel asymmetry and, for small deficits, the optimal exponent $(n+4)/2$ for the normalized Hausdorff distance. All three exponents are sharp even among smooth strictly convex domains whose principal curvatures lie in $[1/2,2]$. In particular, the quadratic Fraenkel-asymmetry estimate proposed by Brasco and De Philippis (2017, Open Problem 7.65) fails. We also determine the optimal stability exponents under higher-order radial regularity bounds.

math.AP↗
arXiv2026

Beyond Correctness: Attribution of Natural-Language Documentation for Formal Libraries

Formal languages are widely used in software engineering to specify and verify program properties. Because their mathematical notation and logic are hard to read, developers often rely on natural-language documentation to understand the functions in formal libraries. Yet a documentation statement can be provably true and still hold after the function's behavior changes. We call a statement's dependence on the function's specific behavior attribution, and we give a method to check it. We design and implement DocAuth, a documentation-generation method that uses attribution to decide what to publish and attaches to each published sentence the formal property recovered from it, together with checkable evidence. On 150 functions from the Community Books library of the ACL2 theorem prover, three large language models produce proved but non-attributable properties for about a third of the functions. DocAuth matches or exceeds the attribution rate of a mutation-based filter while delivering 37% more attributable sentences. Compared with a pipeline that checks truth alone, it raises the share of published sentences that a second model confirms as attributable from 82.9% to 94.7%. These results show that provable truth does not guarantee that documentation reflects what a function actually does. Attribution complements truth as a criterion for generating and evaluating documentation for formal libraries.

cs.SE↗
arXiv2026

OPIS: An Input-Grounded Benchmark for Multi-Object Memory in Video World Models

Video world models must preserve the visual state of the world over time, but existing evaluation protocols often rely on generated histories, video reference, or selected revisit viewpoints that can confound the assessment of a model's true memory capability. To address this, we introduce OPIS, an input-grounded benchmark that strictly anchors the assessment to a fixed set of object instances from the initial observation for evaluating multi-object memory in video world models. The OPIS dataset comprises 500 cases across real-world, embodied-robotic, and game-world domains, providing dense object-level annotations for 12,672 rigid, articulated, and deformable instances. Our object-centric evaluator combines association and explicit visibility reasoning to hierarchically measure Object (O) Presence (P), Identity (I), and Structure (S), utilizing static or dynamic evaluation tracks based on object kinematics. Across eight image-to-video or camera-conditioned world models, our proposed OPIS scores range from 48.65 to 56.01. As the reference inventory grows from less than 20 to more than 40 objects, the Presence, Identity, and Structure scores show an overall decline, with the average Identity score falling from 40.22 to 23.11. The results demonstrate that preserving the particular object instances in the input is considerably harder than generating plausible visual elements.

cs.CV↗
arXiv2026

Extremals for Sharp Trace Gagliardo--Nirenberg--Sobolev Inequalities on Convex Cones

Zugmeyer~\cite{Zug} established sharp trace Gagliardo--Nirenberg--Sobolev inequalities on convex cones for $a\geq n>p>1$. We classify all nonzero real-valued extremals of these inequalities. Let the cone be the epigraph of a finite convex function that is positively homogeneous of degree one, and let $H$ be any norm. We prove that equality holds precisely for functions of the form $A H(x+te+\ell)^{-(a-p)/(p-1)}$, where $A\in\mathbb{R}\backslash\{0\}$, $t>0$, $e=(0,\ldots,0,1)$, and $\ell$ belongs to the largest vector subspace contained in the cone. The result includes the critical Sobolev trace inequality at $a=n$, as well as nonsmooth norms and cones, half-spaces, and cones containing lines.

math.AP↗
arXiv2026

Sharp quantitative stability of the Hong--Krahn--Szegő inequality

We establish sharp quantitative stability estimates for the Hong--Krahn--Szegő inequality for arbitrary open sets of positive finite measure in $\mathbb R^d$, $d\ge2$. In dimensions $d\ge3$, the optimal stability exponent is $(d+2)/2$, while in dimension two the sharp stability modulus is $t^2/\log(e/t)$. These results improve the quantitative stability estimate of Brasco and Pratelli~\cite[Theorem 3.5]{BP} and resolve Open Problem 7.56 of Brasco and De Philippis~\cite[Open Problem 7.56]{BDP}.

math.AP↗
arXiv2026

When Seeing Is Not Believing -- A Benchmark for Search-Grounded Video Misinformation Detection

Video misinformation increasingly operates at the semantic and evidential level: authentic footage may be selectively edited, temporally reordered, spliced across sources, or augmented with AI-generated content to construct false narratives. Such evidence-dependent manipulations cannot be reliably verified from the input video alone, because the missing, reordered, replaced, or recontextualized evidence lies outside the video itself. We introduce \textbf{EVID-Bench}, a benchmark for search-grounded video misinformation detection, where a system must search the open web for related videos and identify what information is false through cross-video comparison. EVID-Bench comprises 222 videos spanning 9 manipulation types across 3 categories: AI generation, single-source editing, and multi-source editing. All samples are verified to be undetectable by frontier models through visual inspection alone. We evaluate nine frontier multimodal models using a retrieval-augmented verification baseline. The best system achieves only 61.43\% point-level accuracy and 43.24\% video-level accuracy, while AI-generated manipulations remain especially challenging. Error analysis reveals recurring challenges: models fixate on irrelevant anchors, misattribute synthetic content to editorial splicing, and terminate search prematurely before fully explaining the manipulation.

cs.CV↗
arXiv2026

Improved Sobolev and Moser--Trudinger--Onofri Inequalities under Constraints: Sharpness and Quantitative Remainders

In this paper, we study improved Sobolev and Moser--Trudinger--Onofri inequalities on the sphere under moment constraints, with particular emphasis on sharp leading constants and quantitative remainder terms. For the Sobolev inequality, we sharpen the almost sharp inequality of Hang and Wang [30] (Int. Math. Res. Not., 2022) to a sharp one at $p=2$ under vanishing moment constraints of orders $m=1,2,3$. We also prove a general failure principle: if the unconstrained sharp Sobolev inequality fails, then the corresponding sharp improved inequality under moment constraints fails as well. For the Moser--Trudinger--Onofri inequality, we sharpen the almost sharp inequalities of Chang and Hang [10] and Hang [29] (Comm. Pure Appl. Math., 2022) to a sharp improved inequality at the critical order $s=\frac{n}{2}$ under arbitrary finite-order moment constraints. We further investigate the best remainder constant. In contrast with the two-dimensional first-moment case of Gui and Moradifam [28] (Invent. Math., 2018), where the best remainder vanishes, we show that it is positive in several higher-dimensional regimes. We also obtain quantitative remainder estimates for general almost sharp improved Sobolev and Moser--Trudinger--Onofri inequalities, which in turn yield sharp inequalities at different scales. We further extend recent second-moment refinements to higher dimensions and higher orders. Finally, we obtain analogous results on sharpness, non-sharpness, and quantitative remainder estimates for improved Sobolev-type inequalities under symmetry constraints on general smooth closed manifolds.

math.AP↗
arXiv2026

ECHO: Efficient Causal Self-Recursion Beyond Auxiliary Multi-Token Prediction

Existing Multi-token prediction (MTP) methods are auxiliary by design: the multi-horizon loss mostly trains modules attached to the backbone, and each added horizon costs another module and another pass. We propose Efficient Causal Self-Recursion (ECHO), which removes both costs. The backbone's own final layer predicts every horizon, so the supervision lands on the parameters that generate the next token, and Dynamic Depth Sampling (DDS) keeps deeper horizons affordable by drawing the training depth with probabilities that decay along depth. Future prediction thus becomes part of the autoregressive objective rather than a task attached to it, and the gains follow on both sides of it. In pretraining, added depth converts into downstream accuracy where auxiliary modules saturate, concentrating on reasoning and code at half the MTP training overhead of training every depth. At inference, the pretrained model is its own drafter: it drafts beyond the depth it was trained for, and keeps its draft-target alignment on the open-ended, high-entropy generation where post-hoc drafters lose theirs. The advantage survives a doubled token budget, a repeated pretraining run, and supervised finetuning.

cs.CL↗
arXiv2026

Towards Unified World Models for Visual Navigation via Memory-Augmented Planning and Foresight

Enabling embodied agents to imagine future states is essential for robust and generalizable visual navigation. Yet, state-of-the-art systems typically rely on modular designs that decouple navigation planning from visual world modeling, which often induces state-action misalignment and weak adaptability in novel or dynamic scenarios. We propose UniWM, a unified, memory-augmented world model that integrates egocentric visual foresight and planning within a single multimodal autoregressive backbone. UniWM explicitly grounds action selection in visually imagined outcomes, tightly aligning prediction with control. Meanwhile, a hierarchical memory mechanism fuses short-term perceptual cues with longer-term trajectory context, supporting stable and coherent reasoning over extended horizons. Extensive experiments on four challenging benchmarks (Go Stanford, ReCon, SCAND, HuRoN) and the 1X Humanoid Dataset show that UniWM improves navigation success rates by up to 30%, substantially reduces trajectory errors against strong baselines, generalizes zero-shot to the unseen TartanDrive dataset, and scales naturally to high-dimensional humanoid navigation. These results position UniWM as a principled step toward unified, imagination-driven embodied navigation. The code and models are available at https://github.com/UWMILab/UniWM.

cs.AI↗
arXiv2026

Interior Regularity for Stable Solutions of Semilinear Elliptic Equations with Sign-Changing Nonlinearities

We investigate interior regularity for stable solutions of the semilinear elliptic equation $-Δu=f(u)$ in an open set $Ω\subseteq \mathbb R^n$, where $f$ is locally Lipschitz and may change sign. We establish sharp interior Hölder estimates for stable classical solutions in every dimension $n\le 9$, thereby fully resolving a question raised by Cabré concerning interior regularity without the nonnegativity assumption on $f$. For $n\le 5$, we also give a short alternative proof of the corresponding Hölder regularity result due to Peng, Zhang, and Zhou. We further establish higher-integrability estimates for the gradient and $L^1$ bounds for the Hessian in all dimensions. For $n\ge 10$, we obtain BMO and Morrey estimates, extending related results previously established for nonnegative nonlinearities. When $f$ is convex, we extend the interior regularity estimates to stable weak solutions through an approximation theorem. As applications, we prove Liouville-type theorems for entire stable solutions of equations with sign-changing nonlinearities across all dimensions.

math.AP↗
arXiv2026

When Non-Adversarial Noise Hurts Learning: Benchmarking Skill Self-Evolution in Agents

Large language model agents increasingly improve by distilling experience into reusable skills, but errors in that experience can persist in learned rules and affect later tasks. Existing evaluations largely examine clean learning or perturbations encountered after skills are frozen, leaving robustness during self-evolution insufficiently understood. We introduce RSE-Bench, to our knowledge the first benchmark for evaluating skill self-evolution under ordinary, non-adversarial noise. It spans four domains and six benchmarks, with controlled perturbations to instructions, environmental evidence, execution records, and outcome--record pairings that preserve the underlying tasks and native scoring rules. The protocol separates noise during learning from input-side noise at evaluation, testing frozen skill stores on held-out tasks. We also introduce \textsc{RSEClaw}, a method-agnostic framework supporting consistent noise injection and process diagnostics of actual exposure and skill updates. Across three methods and two backbones, noisy learning retains only $25.5\%$--$42.1\%$ of the aggregate gain achieved by clean self-evolution, depending on the noise type. Perturbations confined to execution records and outcome--record pairings cause average losses comparable to input-side noise, despite leaving the original executions and evaluations intact. Diagnostics further identify differences in candidate skill text that coexist with small changes in mean held-out performance.

cs.LG↗
arXiv2026

Three-Color Free-Flood-It on Fixed-Height Grids Is Polynomial-Time Solvable

We give a deterministic algorithm for \textsc{Free-Flood-It} on rectangular grids $P_k\square P_n$ with at most three colors. For every fixed height $k$, it computes the minimum number of moves and an optimal sequence in $N^{O(k^2)}$ time, where $N=kn$. This resolves the previously open three-color case on complete $3\times n$ boards. The proof uses a representation of flooding strategies as paintings by connected regions. We show that an optimal painting can be chosen so that its regions form a rooted tree with strong restrictions on the colors along ancestral paths. These restrictions make the part of the tree visible in any fixed-size connected window admit only polynomially many descriptions. The remaining common ancestors may form an arbitrarily long chain. We retain that chain on a stack and use a finite context-free recurrence to minimize the total painting cost without enumerating all possible stack contents. Connectivity information at the boundary ensures that the resulting local descriptions assemble into one valid global painting. The argument also applies to graphs supplied with an ordering into connected layers of bounded size, provided every edge lies within one layer or joins consecutive layers. A family of three-row boards shows that the unbounded ancestor chains handled by the algorithm are necessary even for optimal paintings. For any fixed height and palette size, we also give a deterministic EPTAS and a randomized sampling variant with an explicit failure-probability bound. These approximation results use a separate algorithm with single-exponential dependence on the move budget.

cs.CC↗
arXiv2026

ShotFinder: Imagination-Driven Open-Domain Video Shot Retrieval via Web Search

In recent years, large language models (LLMs) have made rapid progress in information retrieval, yet existing research has mainly focused on text or static multimodal settings. Open-domain video shot retrieval, which involves richer temporal structure and more complex semantics, still lacks systematic benchmarks and analysis. To fill this gap, we introduce ShotFinder, a benchmark that formalizes editing requirements as keyframe-oriented shot descriptions and introduces five types of controllable single-factor constraints: Temporal order, Color, Visual style, Audio, and Resolution. We curate 1,210 high-quality samples from YouTube across 20 thematic categories, using large models for generation with human verification. Based on the benchmark, we propose ShotFinder, a text-driven three-stage retrieval and localization pipeline: (1) query expansion via video imagination, (2) candidate video retrieval with a search engine, and (3) description-guided shot localization. Experiments on multiple closed-source and open-source models reveal a significant gap to human performance, with clear imbalance across constraints: temporal localization is relatively tractable, while color and visual style remain major challenges. These results reveal that open-domain video shot retrieval is still a critical capability that multimodal large models have yet to overcome.

cs.CV↗
arXiv2026

Broadband Purcell Filter for Fast Superconducting Qubit Reset and Readout

Rapid reset and readout of qubit states are essential for quantum error correction, yet accelerating these operations through stronger coupling to a dissipative environment inevitably increases qubit decay via the Purcell effect. Here we present a broadband Purcell filter that decouples the reset and readout paths, enabling both operations to be independently optimized without compromising qubit coherence. The filter employs two engineered notches - an intrinsic notch and a bandstop notch - to provide broadband Purcell protection, together with an additional reset stub that creates a reset mode below the protected band. To enable fast reset while suppressing filter-mediated interactions between qubits, we couple each qubit to a dedicated reset resonator. We experimentally demonstrate Purcell-limited relaxation times exceeding 1 ms across a 1.2 GHz bandwidth, simultaneously with 500 ns readout without a Josephson parametric amplifier and 100 ns reset with 99.6% efficiency. The reset resonator is designed with a deliberate kappa-chi mismatch, which suppresses photon-shot-noise-induced dephasing by a factor of 70 compared to the readout resonator. Our work provides a scalable hardware solution that resolves the traditional trade-off between fast qubit operations and qubit protection, advancing the prospects for fault-tolerant quantum computing.

quant-ph↗
arXiv2026

Why Does Weak-OOD Help? A Further Step Towards Understanding Jailbreaking VLMs

Large Vision-Language Models (VLMs) are susceptible to jailbreak attacks: researchers have developed various attack strategies that bypass the safety mechanisms of VLMs. Among these approaches, jailbreak methods based on the Out-of-Distribution (OOD) strategy have garnered widespread attention due to their simplicity and effectiveness. This paper further advances the understanding of OOD-based VLM jailbreak methods. We show that mild OOD manipulations can achieve stronger jailbreak performance than both clean inputs and overly strong perturbations, a non-monotonic pattern we define as "weak-OOD". We explain this phenomenon through a trade-off between two dominant factors: input intent perception and model refusal triggering. Our evidence suggests that these two factors respond differently to OOD manipulations, which is consistent with a discrepancy between broad pre-training robustness and narrower safety alignment. Building on this insight, we draw inspiration from optical character recognition (OCR) capability enhancement---a core task in the pre-training phase of mainstream VLMs. Leveraging this capability, we design JOCR (Jailbreak via OCR-Aware Embedded Text Perturbation), a practical OCR-readable extension of embedded-text jailbreaks that achieves the best average ASR among evaluated baselines. Code is available at GitHub: https://github.com/Yuxuan2003/weak-ood-jailbreak.

cs.CR↗
arXiv2026

Revisiting Lossy Verification in Speculative Decoding: Mechanisms, Trade-offs, and Failure Modes

Speculative Decoding (SD) accelerates large language model inference by allowing a lightweight draft model to propose tokens that are subsequently verified in parallel by a larger target model. Recent approaches introduce lossy verification schemes to further improve efficiency by relaxing strict distributional matching. Yet such relaxation silently rewrites the decoding distribution, and the resulting acceleration can come at the cost of unstable, sometimes severely degraded generation quality. In this work, we present a principled analysis of the distributions induced by lossy verification methods. We show that many seemingly distinct approaches differ only superficially and can be unified into two categories: truncation-based verification and collaborative verification. We further construct a diagnostic evaluation framework across curated benchmarks. For truncation-based methods, we identify a fundamental pitfall-performance can degrade significantly compared to the true truncation sampling baseline due to distributional distortion. For collaborative verification, we reveal that well-designed relaxation principles, namely overshoot suppression and supervision quality, matter far more than the linear interpolation between draft and target. Our code is available at https://github.com/ZhouYuxuanYX/Fast-HSD.

cs.CL↗