arXiv subjects

Cheng Zhang

Publications and source records attributed to Cheng Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Reading-Level Verification of Kanji Words in Japanese Financial Speech Synthesis

Few-shot voice cloning makes it cheap to produce daily Japanese financial commentary with several synthetic speakers, but kanji words with more than one reading, such as the word for "market" (shijo vs. ichiba) or "overall trend" (taisei vs. oze), are still misread. A common quality gate transcribes each synthesized sentence with automatic speech recognition (ASR), converts both the input and the transcript to kana, and regenerates on mismatch. We show that this gate is unreliable in both directions: when the audio contains a misreading, the ASR language model frequently writes the expected kanji back, hiding the error, and when the audio is correct, ordinary recognition errors trigger needless regeneration. We propose a reading-level check that scores the expected reading and its confusable alternatives with a connectionist temporal classification (CTC) hiragana recognizer and flags a word when an alternative explains the audio better. On 540 utterances from three voice-cloned production voices covering 30 financial terms, with misreadings injected under controlled conditions, the transcript-based gate detected 78.9% of misreadings at a 44.4% false-alarm rate, whereas the proposed check detected 91.7% at 5.0% (AUC 0.990). On natural kanji input, it caught a front-end error that the transcript-based gate missed for one voice, and reduced regeneration requests from 38.9% to 6.1% of sentences.

cs.SD↗

arXiv2026

Persistence of Bounded Rationality in Financial Markets: An Evolutionary Game-Theoretic Model

Classical theory predicts that irrational traders lose money to rational traders and are eventually driven out of the market. Yet noise trading, herding, and heuristic trading persist even in highly developed markets. We model a financial market as an ecology of two trading-behavior types, relatively rational (R) and boundedly rational (N), whose composition evolves under replicator dynamics according to relative performance. Each type's fitness depends on the share of the other type and on frictions. When the interaction is complementary, that is, rational traders profit from the mispricing created by boundedly rational traders while the latter benefit from the liquidity and price correction supplied by the former, and when the difference in frictions is moderate, a unique interior composition exists and is globally asymptotically stable: bounded rationality persists endogenously. When the interaction is antagonistic, the same point becomes a tipping threshold between two one-type regimes. Comparative statics show that lowering the costs of disciplined trading, as algorithmic and AI-based trading does, shifts the composition toward rational trading continuously but does not eliminate bounded rationality unless the cost advantage exceeds an explicit threshold.

cs.GT↗

arXiv2026

Sharp Bounds for Toral Eigenfunctions and Threefold Additive Energy on Circles

For the lattice circle $P_n=\{z\in\mathbb{Z}^2:|z|^2=n\}$, we prove that the threefold additive energy $J_3(P_n)\le C|P_n|^3$, with an absolute constant $C$. This establishes the optimal bound suggested by Krishnapur--Kurlberg--Wigman and Bombieri--Bourgain. We also prove a uniform weak $L^6$ bound for $L^2$-normalized eigenfunctions on the square torus, and hence uniform $L^p$ bounds for every $1\le p<6$. This extends the classical $L^4$ bounds of Cooke and Zygmund and establishes Bourgain's conjecture in this range. Applications include mean-square convergence for $L^{6/5}$ observables and a logarithm-free remainder in the nodal length variance of arithmetic random waves.

math.AP↗

arXiv2026

Reuse, Rewrite, or Reject: Evidence-Constrained Trace Reuse for Repeated Mobile GUI Control

Mobile GUI agents frequently perform repetitive tasks with similar interaction procedures. Reusing previous executions can reduce redundant LLM inference and execution overhead, but variations in user instructions and execution environments may invalidate previously successful actions. We systematically characterize these variations into six representative categories and observe that such changes do not necessarily invalidate entire procedures: individual transitions can remain reusable when validated against the current task context and interface state. Based on this insight, we propose graph-text templates for selective execution-trace reuse. The graph captures UI states and action transitions, while the text encodes task intent, procedural structure, and typed slots for variable content. A record-template-replay framework constructs templates from verified executions and leverages the current instruction and live UI state to reuse, adapt, or discard individual transitions. Given an execution trace, our framework reduces LLM inference token consumption by approximately 80% on average across five base agents on AndroidWorld, with most agents achieving at least 1.5-fold execution speedup over LLM-only execution. In real-device evaluations on Android and HarmonyOS, our framework achieves action reuse ratios of 70.7-79.8% while maintaining over 2-fold execution speedup in most evaluated configurations.

cs.SE↗

arXiv2026

Sharp Bounds for Nodal Sets of Steklov Eigenfunctions on Smooth Surfaces

We prove a sharp linear upper bound for the interior nodal length of Steklov eigenfunctions on smooth compact Riemannian surfaces with boundary, establishing the upper bound predicted by the Steklov analogue of Yau's conjecture in dimension two. On each fixed surface, we also show that the nodal length divided by the logarithm of the eigenvalue tends to infinity, uniformly over the eigenfunctions. We construct examples to show this divergence can be arbitrarily slow. These examples rule out every general lower bound by a positive power of the eigenvalue, in contrast with the linear lower bound on analytic surfaces by Polterovich-Sher-Toth. Moreover, boundary zero counts and both interior and boundary nodal domain counts satisfy sharp linear bounds for large eigenvalues. We also prove density estimates for boundary zeros on at least one boundary component and construct a fixed analytic cylinder on which another component remains free of boundary zeros along an unbounded sequence of eigenvalues.

math.AP↗

arXiv2026

Not Another Text Benchmark: Putting the "Visual" Back in Visual Question Answering for Large Video Models

Large video models have exhibited impressive performance on a wide range of visual question answering tasks, owing to the rise of powerful, pretrained text and vision encoders. The usefulness of such models have also been demonstrated on a wide range of benchmarks, with an important caveat - the dominant approach in these benchmarks evaluates multiple choice reasoning via text options. This is a natural way to test text-based reasoning in these models, and has led to significant insights regarding model behavior in the community. In this work, we ask a different question - what happens when the evaluation modality is visual, rather than text? We introduce three new vision-centric evaluation benchmarks in temporal frame retrieval, video future prediction, and causal memory distortion, all designed around evaluating visual understanding capabilities in large video models. Our approach complements the existing approaches to evaluate video understanding in frontier models. We show that current frontier models exhibit significant weakness when attempting to reason through visual queries, rather than text. We conclude with an extended analysis section that provides pointers for future improvements in visual understanding for large video models.

cs.CV↗

arXiv2026

ScopeMem: Latent Memory That Learns Where to Revise Frozen GUI Agents

Latent memory is gaining ground as a lightweight history supplement for GUI agents. It supplies history to a frozen vision-language model via prefix vectors trained with step-wise cross-entropy (CE). While resulting gains are naturally attributed to encoded history, our controlled experiments show they primarily stem from action revision. The prefix learns to alter the model's default actions. However, this revision is unscoped because step-wise CE pushes outputs toward reference actions at every step without explicitly preserving correct decisions. Consequently, it repairs errors but also breaks 8-13% of initially correct answers. Overall accuracy metrics hide these losses. We therefore propose ScopeMem, a latent-memory method for scoped revision. Its encoder conditions prefix generation on the model's initial answer and is trained to correct errors while preserving correct decisions. On the controlled ablation set, ScopeMem reduces the share of broken correct decisions to 5.6% while repairing more errors than any evaluated CE-trained prefix. On AndroidControl, GUI-Odyssey, and zero-shot AITW, ScopeMem outperforms all evaluated prompting and latent-memory baselines, enabling a frozen Qwen3-VL-8B to match or surpass GUI-specialized agents and Qwen3-VL-32B.

cs.AI↗

arXiv2026

Denoising Implicit Variational Inference

Semi-implicit variational inference (SIVI) expands the representational capacity of variational families, but optimizing the Evidence Lower Bound (ELBO) remains challenging because the marginal score of the semi-implicit distribution is generally intractable. Unbiased Implicit Variational Inference (UIVI) and related methods address this difficulty through reparameterized ELBO gradients, but existing approaches estimate the required score indirectly through reverse-conditional sampling or approximation, which can be computationally costly and unstable. We propose Denoising Implicit Variational Inference (DIVI), which learns the marginal score directly via denoising score matching. The learned score is used as a plug-in estimator in the pathwise ELBO gradient, replacing inner reverse-conditional sampling with a score-network evaluation. We analyze DIVI as inexact stochastic gradient ascent and show, under stated assumptions, that the averaged stationarity measure is controlled by the usual optimization term and the average score-matching error. Empirically, we evaluate DIVI on both synthetic distributions and a variety of real-data Bayesian inference tasks. The results show that DIVI improves over the current UIVI baselines while reducing the cost of ELBO-gradient estimation.

stat.ML↗

arXiv2026

Control Variate Design for Score-Function Gradient Estimators in Importance-Weighted Variational Inference

Importance-weighted variational inference approximates intractable densities by optimizing multi-sample Monte Carlo objectives whose tightness increases with the number of importance samples $N$. When reparameterization is unavailable, optimization relies on score-function (REINFORCE) gradient estimators. In this paper, we provide a comprehensive analysis of score-function gradient estimators for importance-weighted objectives by characterizing their asymptotic expectation and variance. We establish that standard variational inference for Monte Carlo objectives (VIMCO) gradient estimators suffer from a vanishing signal-to-noise ratio (SNR) as $N$ increases, which arises from a suboptimal control variate choice and can impair optimization. To address this issue, we derive a general mechanism for constructing VIMCO gradient estimators that avoid SNR collapse. We identify the High-Effective Sample Size optimal variance importance sampling estimator as belonging to this category of estimators and establish its $\sqrt{N}$ SNR scaling. This estimator does not, however, have zero variance at optimality and achieving its $\sqrt{N}$ SNR rate may require prohibitively large $N$ when the importance weights are degenerate. We overcome these limitations by introducing VIMCO gradient estimators that attain zero variance at optimality and support importance weight tempering. Experiments with discrete latent-variable and state-space models validate the theoretical predictions and illustrate the benefits of our estimators.

stat.ML↗

arXiv2026

MemCut: Every Secret Leaves a Trail in the Long-Term Memory of LLM Agents

Long-term memory has become a crucial component of large language model (LLM) agents, with most research focusing on retrieving relevant information accurately and efficiently. However, a dependable memory system must also forget sensitive user data upon request. To evaluate this capability, we introduce MemCut, a systematic benchmark for user-requested forgetting. Its 456 hand-designed secrets in long-horizon conversations have fully annotated evidence trails to test recovery after deletion, including from messages that never state the secrets directly. We propose two recovery metrics. Native recovery measures reader success through the system's retrieval, whereas residual recovery measures the same reader's success when directly given surviving objects traced to annotated disclosures or evidence. We establish three reference deletion baselines to test how far existing retrievers and write models support forgetting across five memory designs and four LLMs. Disclosure removal reduces native recovery far more than residual recovery. LLM-guided deletion of ten additional turns from twenty retrieval candidates leaves 39-76% residual recovery. Subset sweeps (144 facts, three systems, two models) show that widening the pool at a fixed budget reduces residual recovery more than deleting all original candidates. These sweeps identify candidate coverage as a bottleneck in the reference baselines.

cs.AI↗

arXiv2026

A New Bound for the Gauss Circle Problem

Let $E(X)=\#\{(a,b)\in\Z^2:a^2+b^2\le X\}-πX$. We prove that $E(X)=O_\varepsilon(X^{209/670+\varepsilon})$ for every $\varepsilon>0$, uniformly for real $X\ge4$. This breaks the $5/16$ threshold suggested by Huxley as a possible limit of the Bombieri--Iwaniec method at the time. The proof introduces two refinements: a cone distribution estimate that preserves separate fourth- and sixth-power contributions in the double large sieve, and an explicit inversion of Taylor-coefficient constraints that cancels the linear error in a pair-counting argument. These refinements, together with the Robert--Sargos and Watt spacing estimates and the Vinogradov mean value theorem, yield the improvement.

math.NT↗

arXiv2026

FastEEP: Fast Intra-Instance Elastic Scaling for Expert-Parallel MoE Serving

Large-scale expert parallelism (EP) has become the standard deployment configuration for Mixture-of-Experts (MoE) models in production. Meanwhile, the highly dynamic nature of LLM workloads requires providers to continuously adjust serving capacity to match fluctuating demand. Scaling MoE serving cannot be efficiently achieved by simply adding or removing serving instances; instead, it often requires resizing the EP group within an existing instance, which we refer to as intra-instance scaling.Existing inference engines primarily rely on instance-level scaling, which destroys and reconstructs entire serving instances, incurring substantial scaling overhead. We present FastEEP (Fast Elastic Expert Parallel), a fast intra-instance scaling system for MoE serving based on the principle of Partial Reuse. FastEEP reuses existing runtime state across worker execution, GPU memory, and communication by separating reusable execution structures from the resources activated by the current EP configuration. This separation enables coordinated resource reclamation and reconfiguration while preserving costly-to-reconstruct state, allowing fast EP scaling without full instance teardown and reconstruction. Our evaluation shows that FastEEP achieves 4.5x--11.7x faster scaling than the existing vLLM-based approach across different models, enabling efficient, fine-grained scaling of MoE serving.

cs.DC↗

arXiv2026

Critical quantum sensing via backaction-evading continuous measurement

Continuous monitoring of open quantum-optical systems offers a promising route towards quantum-enhanced sensing. In such sensing schemes, the ultimate precision limit is dictated by the quantum Cramér-Rao bound. However, reaching this fundamental bound with established continuous measurement techniques has remained an outstanding challenge. Here, we present an approach tailored to open quantum-optical sensors operating near dissipative critical points. The key insight of our approach is to exploit a synergy between the intrinsic critical dynamics and the unavoidable quantum measurement backaction of continuous homodyne detection. By carefully aligning the homodyne phase to the squeezing angle of the sensor state, quantum backaction can be evaded in the critical dynamics, thereby significantly boosting the attainable precision. We demonstrate the broad applicability of this approach across several sensor models and for both single- and multi-parameter estimation, showing that it consistently narrows the gap toward the ultimate precision bound.

quant-ph↗

arXiv2026

Cubic moments of eigenfunctions on surfaces

We identify Gaussian curvature as a local mechanism for cubic cancellation of Laplace eigenfunctions on Riemannian surfaces, without arithmetic or dynamical assumptions. For $L^2$-normalized local solutions at frequency $λ$, cubic integrals against smooth compactly supported weights tend to zero whenever the curvature is nonzero almost everywhere where the weight is nonzero. Finite-order nondegeneracy yields sharp polynomial decay rates. In particular, on closed surfaces with nowhere-vanishing curvature, we obtain the rate $O(λ^{-1/6})$, attained by spherical harmonics on the round sphere. This proves the cubic case of the Gaussian moment conjecture for all normalized real eigenfunctions on closed negatively curved surfaces. The local proof rests on an exact identity relating the reduced phase at a Fermat point to Gaussian curvature. On complete hyperbolic surfaces with positive systole, we also prove global $O((\logλ)^{-3/2})$ decay for $L^2$-normalized eigenfunctions and smooth weights with bounded covariant derivatives through order two, without assuming finite area.

math.AP↗

arXiv2026

Complex Magnetic Phases and Unconventional Skyrmion Thermodynamics in Magnetically Intercalated Cr$_{1+x}$Te$_2$Compounds

Chemical intercalation provides a powerful route for tuning the electronic and magnetic properties of transition metal dichalcogenides (TMDs), enabling the emergence of magnetic phases and phenomena that are otherwise inaccessible in their pristine forms. Here, we investigate the structural, electronic, and magnetic properties of magnetically intercalated Cr$_{1+x}$Te$_2$ through a combination of scanning tunneling microscopy, angle-resolved photoemission spectroscopy, magnetization, electrical transport, and Lorentz transmission electron microscopy measurements. We uncover a rich magnetic phase diagram comprising paramagnetic, in-plane spin-fluctuation, out-of-plane ferromagnetic, and field-induced Néel-type skyrmion phases. In particular, Lorentz transmission electron microscopy directly reveals the formation of Néel-type skyrmion textures slightly below the Curie temperature, with their configurations evolving strongly with temperature and magnetic field. The skyrmions exhibit pronounced thermal hysteresis and coexist with stripe domains, indicative of a complex magnetic energy landscape arising from the competition among disorder-induced local Dzyaloshinskii-Moriya interactions, perpendicular magnetic anisotropy, and dipolar interactions. Our results demonstrate that chemical intercalation can profoundly reshape the magnetic interactions and phase behavior of TMDs, establishing Cr$_{1+x}$Te$_2$ as a versatile platform for exploring tunable magnetism, skyrmion physics, and unconventional thermodynamic phenomena in low-dimensional quantum materials.

cond-mat.mtrl-sci↗

arXiv2026

Synthesizing Ultrasonic Echo Profiles from Text for Open-Vocabulary Silent Speech Recognition

Eyewear that senses articulation with inaudible ultrasound can now recognize silent speech over an open vocabulary, but its accuracy is limited by the cost of training data. We present SoniGAN, which synthesizes the ultrasonic echo profiles of silent articulation from text and uses them as synthetic supervision. A voice-cloned text-to-speech (TTS) model reads a text corpus, and a two-stage generator synthesizes echo profiles from representations of the spoken content of that audio. Pretraining a recognizer on 104 hours of the synthesized data improves the word error rate (WER) of a public open-vocabulary benchmark by 2.3 points. Adding the sessions the corpus discards for a faulty microphone brings the system to 22.2% without any language model. Lexicon-constrained decoding, LLM rescoring, and ensembling further bring it to 14.9%, the first minimally obtrusive silent speech interface to fall inside the error range in which users report no loss of willingness to use silent speech input.

cs.SD↗

arXiv2026

Maximal growth of the Stein-Wainger oscillatory integral

We determine the optimal growth of the Stein--Wainger oscillatory integral for every Denjoy--Carleman class defined by a positive log-convex sequence. Quasianalyticity is equivalent to boundedness in $λ$ for every fixed phase in the class. In the nonquasianalytic case, an explicit formula in the defining sequence gives the optimal order, attained by a single phase for all sufficiently large frequencies. This yields sharp bounds for Gevrey classes, their logarithmic refinements, and classes beyond every Gevrey class. For general smooth phases the growth is $o(\logλ)$, and every prescribed $h(λ)\to\infty$ with $h(λ)=o(\logλ)$ is attained asymptotically along a sequence by one smooth phase. These results answer a question of Wang and Zhang motivated by endpoint eigenfunction restriction.

math.CA↗

arXiv2026

CAG-WAM: Causal Action-Gain World Action Models for Robot Manipulation

World Action Models (WAMs) combine action generation with future prediction, allowing predicted futures to support action evaluation and robot decision making. However, the future associated with a candidate action may contain action-induced and environment-driven changes. Directly evaluating the predicted future may not fully characterize the future change associated with the candidate action. For example, a candidate action that worsens a favorable environmental future may still yield a favorable outcome, while one that improves an unfavorable environmental future may still yield an unfavorable outcome. To analyze this issue, we formulate a Structural Causal Model (SCM) for action evaluation in WAMs. The SCM formalizes how action-induced and environment-driven changes jointly shape future evaluation, and reveals a backdoor path through which environmental evolution can also affect the future evaluation associated with a candidate action. Intervention and backdoor adjustment account for this path, providing a causal basis for characterizing the future change associated with each candidate action. Based on this analysis, we propose Causal Action-Gain World Action Model (CAG-WAM). For each candidate action, CAG-WAM predicts an action-conditioned future and an environmental future under the same current observation, and defines their evaluation difference as causal action gain (CAG) for characterizing the future change associated with the candidate action. During training, we incorporate CAG into Group Relative Policy Optimization (GRPO) to guide robot policy learning. During inference, we evaluate the initial action with CAG and, when needed, revise it using predicted future information. Experiments on multiple benchmarks support the effectiveness of CAG-WAM.

cs.RO↗