Search arXiv⌕ Search

arXiv subjects

Yu Zhou

Publications and source records attributed to Yu Zhou.

At least 19 recordsLinked to original sources
arXiv2026

Chain-of-Thought in Neural Code Generation: From and For Lightweight Language Models

Large Language Models (LLMs) have demonstrated remarkable potential in code generation. The integration of Chain of Thought (CoT) reasoning can further boost their performance. However, current CoT methods often require manual writing or LLMs with over 100 billion parameters to generate, impeding their applicability in resource-constrained scenarios. In this study, we investigate lightweight Language Models (lLMs), which are defined to have fewer than 10 billion parameters. Empirically, we find that most lLMs cannot generate high-quality CoTs when prompted by the few-shot method, but can take advantage of high-quality CoTs generated elsewhere to improve their performance in code generation. Based on these findings, we design a novel approach COTTON which can leverage lLMs to automatically generate CoTs for code generation. We synthesize new datasets and conduct extensive experiments on various benchmarks. The results show that the CoTs generated by COTTON outperform the baselines in terms of automated and human evaluation metrics. In particular, the CoTs generated by COTTON boost various lLMs to achieve higher performance gains than those generated by LLMs such as ChatGLM (130B), and are competitive with those generated by gpt-3.5-turbo (175B). Our study also showcases the potential of lLMs in software engineering applications.

cs.SE↗
arXiv2026

OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory

Multimodal large language models (MLLMs) have made rapid progress on visual reasoning, yet many tasks require evidence that is difficult to obtain through internal reasoning alone: our controlled study on 22 benchmarks shows that simply increasing the inference budget of the base model yields only modest gains. Such tasks demand grounded perception, external knowledge, and numerical measurements obtained through heterogeneous tools, whose outputs must be made interpretable to the model and retained across reasoning steps. We introduce OpenVLHarness, a multimodal reasoning harness organized around a capability layer and persistent multimodal memory. The capability layer combines task-level tool interfaces with evidence rendering, providing consistent access to heterogeneous backends and converting their outputs into model-readable observations. Persistent multimodal memory retains the corresponding images, structured data, and web references as addressable artifacts throughout each query. A coding interface executes generated programs over stored data, enabling computations that combine results from multiple tools. Together, these mechanisms support an iterative process of acquiring, inspecting, and composing multimodal evidence. Evaluated on 22 benchmarks with eight open and proprietary backbones, OpenVLHarness improves average scores by 9.7-12.6 points over matched no-tool baselines, including +11.2 with Qwen3-VL-8B, +9.7 with Kimi K3, and +9.9 with GPT-6 Astra. With GPT-6 Luna at medium reasoning effort, OpenVLHarness outperforms the base model, Codex, and Visual Sketchpad at maximum effort by 7.7-8.5 points while incurring at most 37% of their model-token cost.

cs.CL↗
arXiv2026

Flaw-tolerant recycling of waste tires into high-performance rubber

Annually about one billion tires reach the end of life. Rubber from waste tires can be ground into particles and incorporated into a matrix. In practice, the resulting composites are commonly not recycled into new tires, but downcycled into low-value products such as sport surfaces, indoor flooring, and rubberized asphalt. Recycling waste tires into new tires remains a challenge. In particular, such composites typically have low strength, especially when the particle content is high or the particles are large. A single waste tire particle may act as a flaw to initiate macroscopic fracture and knock down the strength of the entire composite. Here we propose the principle of flaw-tolerant recycling. Rather than requiring every particle to be reliable and small, the matrix is designed so that flaw-like particles do not initiate macroscopic fracture. We demonstrate a flaw-tolerant matrix consisting of two intertwined networks: a network of densely entangled and sparsely crosslinked long natural rubber chains, and a percolated network of carbon aggregates. When the flaw-tolerant matrix is incorporated with up to 80 wt% of commercially available waste particles, the resulting composites still maintain strength. The principle of flaw-tolerant recycling may aid the development of recycling waste tires into new tires.

cond-mat.soft↗
arXiv2026

The Holistic Storage of Verb+Up Phrases in Text-based and Audio-based Language Models

One of the most central aspects of language processing is the ability to trade off between stored representations and abstract knowledge: one must retrieve stored representations, but also generate novel ones by applying productive rules. While recent work has examined abstract knowledge in language models, holistic storage has received far less attention. We probe internal representations in both text-based LLMs and an ASR model, testing whether V+up phrasal verbs develop distinct representations as a function of frequency and predictability. All models show evidence of holistic storage driven by frequency and predictability, further supporting usage-based theories of language.

cs.CL↗
arXiv2026

Everyone Was Right: Lexical Representations in Language Models Are Both Abstract and Holistic

Lexical representations can encode both abstract information and information specific to particular contexts. We investigate how these forms of information are organized in language models by decomposing contextual word representations into item, class, and word-class interaction components. Across six pretrained transformers and two linguistic categories, we find that the interaction is consistently present, grows across layers, and contains information associated with both item and class. Unlike the nearly orthogonal item and class effects, the interaction overlaps with both and encodes class information beyond the shared class effect. An interaction also emerges in artificial-language learners even when the input contains only additive structure. Together, these findings show that lexical representations combine abstract information with structure specific to particular word-class combinations.

cs.CL↗
arXiv2026

StepAudio 3 Realtime Technical Report

Realtime spoken interaction demands deep reasoning, prompt responses, and fluid turn-taking. We present StepAudio 3 Realtime, an audio-language foundation model organized around a continuous listen-converse-think-act loop. Deep Perception captures rich acoustic cues to interpret user intent, while Seamless Duplex models synchronized audio streams to handle pauses, backchannels, and interruptions naturally. Crucially, we resolve the tension between deep deliberation and latency via Think-While-Speaking, executing private reasoning in parallel with spoken delivery. In reasoning mode, StepAudio 3 reaches a 73.0 macro average on StepAudioChat. With Think-While-Speaking, it achieves dialogue and reasoning performance comparable to dedicated reasoning models while speaking in real time. Furthermore, an integrated Voice Agent handles asynchronous tool execution without disrupting the dialogue flow. StepAudio 3 Realtime achieves top-tier performance across key dimensions: an exceptional 90.6 on the MMSU benchmark, 98.9 Overall on the Artificial Analysis Full-Duplex Bench, and a 56.0% macro task-success rate on $τ$-Voice.

cs.SD↗
arXiv2026

Finite-event modulation transfer through forward-scattering media

How spatial structure is transferred through a forward-scattering medium depends jointly on pattern scale and propagation distance. Starting from photon exit displacements, we derive modulation transfer while retaining the number, depths, and projected angles of individual deflections, with event occurrences described by a homogeneous Poisson process along the propagation path. The second angular moment yields a leading low-frequency loss proportional to $k^2z^3$, where $k$ is angular spatial frequency and $z$ is thickness. At finite frequency, the model predicts two coupled departures: loss bends below its low-frequency quadratic extrapolation at fixed thickness, and transfer curves separate by thickness at fixed $k^2z^3$. The leading corrections to both effects arise from the same fourth angular moment. For symmetric, nondegenerate projected-angle distributions, the full response further establishes increasing loss with thickness in the same medium at fixed joint variable. Binary-pattern measurements through TiO$_2$ suspensions span five nominal loadings, seven thicknesses, and 20 spatial frequencies, supporting the approximate low-frequency scaling and showing both departures. A Gaussian effective-angle response calibrated only on lower-frequency data predicts the held-out higher-frequency band without refitting, with a logarithmic root-mean-square error of 0.0181 compared with 0.1021 for quadratic extrapolation. The results show that finite-event statistics provide a common physical origin for spectral bending and thickness-dependent departure from low-frequency scaling, linking coarse-pattern transport to finer spatial structure.

physics.optics↗
arXiv2026

Unlocking the potential of polymer networks for high strength through topology

A covalent polymer network is orders of magnitude weaker than individual covalent bonds. This large gap reveals an unrealized potential. The weakness of a network comes not from individual covalent bonds, but from how they are connected. Here we unlock this potential by identifying a topological criterion, involving optimally sparse crosslinks, long chains, and dense entanglements. Optimizing each topological variable alone is insufficient for high strength, yet exceptional strength emerges when they are combined within a narrow regime. We describe the physics of this synergy, and use the criterion to strengthen networks widely used but known to be weak. For example, we prepare a polyacrylamide hydrogel of 90 wt% water, amplifying strength from a commonly reported value of ~0.1 MPa to ~6.5 MPa. The exceptional strength is achieved without sacrificing other properties, such as softness, toughness, and elasticity. The criterion establishes topology as a general design principle for strong polymer networks.

cond-mat.soft↗
arXiv2026

Elastic polymer networks of exceptional strength by deconcentrating tension

The strength of a polymer network is orders of magnitude below that of a polymer chain, because the network concentrates high tension in a small fraction of polymer strands. Here we show that the strength of a network can be greatly amplified by recruiting a larger fraction of strands to bear high tension. We develop approaches to fabricate hydrogels of exceptional strength while maintaining low hysteresis. The strength of the hydrogel increases from ~0.05 MPa for a regular network, to ~1 MPa for a highly entangled network, and further to ~10 MPa for a prestretched interpenetrating network. Similar amplifications of strength are achieved for elastomers. Furthermore, experimental data suggest a scaling relation between strength and strand length. This work provides design principles for creating elastic and strong polymer networks.

cond-mat.soft↗
arXiv2026

Robust quantum key distribution with imperfections in state preparation and characterization

Quantum key distribution (QKD) offers an approach for secure communication over quantum channels, based on the principles of quantum mechanics. However, the inevitable imperfections in QKD systems can introduce critical vulnerabilities, posing significant challenges to the deployment of QKD in terms of both practical security and performance. In this work, we present a comprehensive security analysis of QKD in the practical scenarios involving imperfections in modulation spaces. Specifically, our analysis focuses on the imperfections related to encoding, intensity modulation, and their characterization. We provide a unified framework based on the operator dominance method to overcome these issues and achieve a tight estimation of the secret key rate (SKR). The simulation results show that the SKR in practical scenarios, with parameters characterized from commercial QKD systems, closely matches the SKR under ideal conditions. In the finite case, the maximum transmission distance can reach up to 250 km with total pulse number $N=10^{12}$ and the SKR closely approaches the asymptotic limit up to 250 km with $N=10^{14}$. This validates the robustness of our method against imperfections in state preparation and characterization. Furthermore, the feasibility and robustness of method in measurement-device-independent (MDI) QKD have also been demonstrated. Our analysis not only promotes advancements in enhancing practical security and performance but also highlights the feasibility of deploying QKD for actual applications.

quant-ph↗
arXiv2026

Experimental Measurement-Device-Independent Quantum Key distribution with Flawed State-Preparation over 300 km

Quantum key distribution (QKD) promises theoretically secure communication. However, it encounters challenges in implementation security and performance due to inevitable device imperfections. Since the proposal of measurement-device-independent (MDI) QKD, the critical step toward practical security is to secure QKD with imperfect sources. The source imperfections manifest as state-preparation uncertainty (SPU) in various aspects, e.g., encoding uncertainty, intensity fluctuation, and imperfect vacuum states. Here, we perform an MDI-QKD experiment and achieve both high practical security and superior performance. We address the general form of SPU and guarantee a tight estimation of the secret key rate based on the operator dominance method. We achieve secure key distribution over 303.37 km, which not only represents the farthest distance in experiments involving SPU but also considers the most SPU scenarios. Our experimental results represent a significant step toward promoting practical and secure quantum communication.

quant-ph↗
arXiv2026

Realisation of a Protected Cat-Qutrit Manifold via Engineered Quantum Tunnelling

Engineering quantum tunnelling in phase space has emerged as a viable method for creating a protected logical qubit manifold with biased-noise properties. A promising approach is to combine a Kerr nonlinearity with a multi-photon drive, resulting in a system known as a Kerr parametric oscillator (KPO). In this work, we implement a three-photon KPO and explore its potential as a protected bosonic qutrit. We confirm quantum coherence by demonstrating three-photon Rabi oscillations and performing direct Wigner function measurements that reveal the formation of three-component cat-like states. Crucially, we observe a breathing-like dynamic in phase space, a characteristic feature of driven quantum systems. This dynamic arises from macroscopic temporal interference between the cat-qutrit manifold and the excited states. The frequency of resulting oscillations in the mean photon number provides a direct, time-domain measurement of the energy gap separating the qutrit from the excited states, thereby establishing an experimental hallmark of qutrit manifold protection. Furthermore, we identify a parasitic higher-order pump term as the primary mechanism constraining the mean photon number, highlighting its mitigation as a requisite for maximising protection. Our findings elucidate the basic quantum properties of the three-photon KPO and establish the first step towards its use as an alternative qutrit platform.

quant-ph↗
arXiv2026

Phase-Field Fracture Simulation of Highly Deformable Thin Structures via a Discrete Differential Geometry Framework

Thin elastic flexible structures exploit large geometric deformation to achieve mechanical functionality, making their fracture behavior strongly coupled with the evolving structural configuration. This work presents a unified discrete differential geometry (DDG) phase-field framework for fracture in highly deformable thin structures. Membrane and bending elasticity are formulated from discrete geometric measures on a triangulated midsurface, while the phase field is defined on the same discrete surface to describe onset of crack growth and subsequent propagation. The coupled problem is solved using a staggered scheme, with phase-field irreversibility enforced by an active-set method. A degradation-deletion procedure removes nearly fully failed elements and reconstructs the DDG topology, enabling complete crack opening and substantial post-fracture reconfiguration. For two-dimensional in-plane fracture problems, the DDG predictions agree closely with geometrically nonlinear finite element results in both crack evolution and mechanical response. In three-dimensional tearing, the framework resolves the interaction between out-of-plane deformation and crack propagation, including the experimentally observed convergence and coalescence of initially parallel cracks. Its engineering applicability is further demonstrated using an island--bridge structure representative of flexible electronics. The simulations capture direction-dependent deformation modes, instability and snap-through during loading, and the coupled evolution of structural configuration and fracture. These results establish the proposed framework as an effective tool for investigating geometry-dependent fracture and supporting the damage-tolerant design of flexible thin structures.

cond-mat.soft↗
arXiv2026

Forecasting cosmological constraints from the nonlinear halo density field with Zel'dovich-inspired transform

The Zel'dovich-inspired (ZI) transform is a local nonlinear transform of the density field that can Gaussianize the late-time matter field and recover information from higher-order $N$-point correlation functions through its power spectrum, with the parameter $η$ controlling the weighting of higher-order information. Using Fisher-matrix analyses of halo catalogs from the Quijote and Quijote-PNG simulations, we study the redshift-space monopole power spectrum of the ZI-transformed halo density field. The joint statistic $P_{\rm ZI,0}$, constructed from three representative transforms with $η=1$, $10$, and $+\infty$, yields substantially tighter constraints than the conventional halo power spectrum $P_{\rm h,0}$ on standard cosmological parameters, the summed neutrino mass, and primordial non-Gaussianity amplitudes, while outperforming the marked power spectrum for many parameters. We further calibrate a scale-dependent noise correction for the ZI power spectrum, with an effective noise amplitude that depends on wavenumber, and find that the correction further tightens the constraints. In particular, the noise-corrected $P_{\rm ZI,0}$ improves the constraints on $f_{\rm NL}^{\rm equil}$, $f_{\rm NL}^{\rm ortho}$, $M_ν$, and $n_{\rm s}$ by factors of $5.6$, $5.5$, $5.2$, and $5.8$, respectively, relative to $P_{\rm h,0}$. These results demonstrate the potential of ZI power spectra as a complement to standard two-point statistics for extracting nonlinear cosmological information. Our forecasts may be conservative because the halo catalogs have substantially lower number densities than those expected from forthcoming galaxy surveys.

astro-ph.CO↗
arXiv2026

On the Interpretability of Whisper Encodings Using Sparse Autoencoders

While deep transformer-based models have advanced rapidly, their internal mechanisms remain largely a mystery. Recent work has prioritized understanding text-based transformer models, leaving ASR systems largely unexplored. In order to address this gap, we examine the internal representations of Whisper's encoder using a sparse autoencoder. We find diverse monosemantic features across linguistic and non-linguistic boundaries, spanning a hierarchy from phonetic to semantic representations, and conduct a causal feature-steering campaign across this hierarchy, including cross-lingual steering. We further find that steering is more reliable for higher-level features than lower-level ones, an asymmetry that may reflect redundant encoding of lower-level information. Altogether, this work demonstrates that Whisper's encoder represents a surprisingly rich hierarchy of linguistic information that extends well beyond what is strictly necessary for transcription.

cs.CL↗
arXiv2026

Polymorph-Tunable Spin Texture and Excitonic Structures in Monolayer WSi2P4

Two-dimensional semiconductors that simultaneously possess a direct band gap and strong spin-orbit coupling (SOC) are highly attractive for quantum optoelectronics. Using first-principles GW plus Bethe-Salpeter equation (GW-BSE) calculations, we show that monolayer WSi2P4-an experimentally accessible member of the MSi2X4 family-hosts a direct K-valley gap together with strong SOC. Its three competing polymorphs (alpha, beta and gamma) are all direct-gap semiconductors and are kinetically locked behind 1.3 eV migration barriers. The sigma_h mirror plane of the D3h of alpha and gamma phases enforces a persistent spin texture across the Brillouin zone, weakly modulated near gamma by interband SOC mixing, whereas the polar beta phase (C3v) exhibits Rashba spin splitting. SOC splits the doubly-degenerate lowest bright exciton into dark states and splits the original absorption peak into two peaks (A and B), whose relative brightness is governed by the K-valley conduction-band splitting. alpha-WSi2P4 displays the brightest Peak A because a tiny band crossing of opposite-spin branches can open a spin-allowed radiative channel. These results establish the MSi2P4 family as a phase-tunable platform for the cooperative engineering of spin texture, band splitting, and excitonic brightness within a single material system.

cond-mat.mtrl-sci↗
arXiv2026

Reshaping Rollout Workloads for Asynchronous RL Post-Training on Heterogeneous Accelerators

Reinforcement learning (RL) post-training increasingly relies on long-horizon, multi-turn rollouts. As post-training jobs outgrow a single cluster, rollout pools assembled across clusters introduce hardware heterogeneity. Rollout scheduling must serve two stakeholders: the hardware needs high aggregate decode throughput, while each trajectory needs to finish quickly. The tension arises from the memory-bandwidth-bound nature of autoregressive decoding. A large active batch amortizes weight reads for high throughput but leaves each trajectory a smaller bandwidth share and a longer completion time. The scheduling objective is therefore specialization, letting different workers serve different roles. Heterogeneous hardware further enables this specialization. High-bandwidth accelerators favor long-context work, while cost-efficient accelerators sustain large batches. Workload evolution makes this specialization difficult to sustain, and dynamic reassignment faces a circular dependency because a move's benefit depends on subsequent placement decisions. We present CadenceRL, which bypasses this dependency through structural workload reshaping rather than per-move benefit estimation. Pacing replaces long-context trajectories with shorter ones, providing a structurally positive transformation that sustains large active batches for high throughput. When accumulated staleness demands faster completion, concentration directs the residual long-context tail onto high-affinity workers. Late-bound KV preparation stages accumulated prefixes before a destination is selected. On heterogeneous rollout pools, CadenceRL improves decode throughput by up to 48% and reduces P95 trajectory latency by up to 64%. Adding high-bandwidth accelerators reduces tail latency, while adding cost-efficient accelerators increases throughput, without manual routing configuration.

cs.DC↗
arXiv2026

Chronicles-OCR: A Cross-Temporal Perception Benchmark for the Evolutionary Trajectory of Chinese Characters

Vision Large Language Models (VLLMs) have achieved remarkable success in modern text-rich visual understanding. However, their perceptual robustness in the face of the continuous morphological evolution of historical writing systems remains largely unexplored. Existing ancient text datasets typically focus on isolated historical periods, failing to capture the systematic visual distribution shifts spanning thousands of years. To bridge this gap and empower Digital Humanities, we introduce Chronicles-OCR, the first comprehensive benchmark specifically designed to evaluate the cross-temporal visual perception capabilities of VLLMs across the complete evolutionary trajectory of Chinese characters, known as the Seven Chinese Scripts. Curated in collaboration with top-tier institutional domain experts, the dataset comprises 2,800 strictly balanced images encompassing highly diverse physical media, ranging from tortoise shells to paper-based calligraphy. To accommodate the drastic morphological and topological variations across different historical stages, we propose a novel Stage-Adaptive Annotation Paradigm. Based on this, Chronicles-OCR formulates four rigorous quantitative tasks: cross-period character spotting, fine-grained archaic character recognition via visual referring, ancient text parsing, and script classification. By isolating visual perception from semantic reasoning, Chronicles-OCR provides an authoritative platform to expose the limitations of current VLLMs, paving the way for robust, evolution-aware historical text perception. Chronicles-OCR is publicly available at https://github.com/VirtualLUOUCAS/Chronicles-OCR.

cs.CV↗