arXiv subjects

Yichi Zhang

Publications and source records attributed to Yichi Zhang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation

Reinforcement learning can turn one language model into several specialists, each excellent at a single skill such as mathematics, coding or following instructions, but users need one model with all of these skills. Multi-teacher on-policy distillation (MOPD) merges them by letting the specialists teach one student: the student answers each prompt, and the specialist for that prompt's domain gives feedback on every token. This routing decides which specialist teaches, but not how strongly its feedback moves the shared student. In Qwen3.5 models at three sizes, we find that MOPD's student does not beat one taught by the best single specialist and gains little of the mathematics specialist's advantage. The feedback is unbalanced: instruction-following feedback is several times more spread out than mathematics feedback and dominates the student's updates. We propose Domain-Normalized MOPD (DN-MOPD), which keeps the routing and rescales each domain's feedback by its measured spread. On six public benchmarks, DN-MOPD improves the average score over MOPD at every size, across three random seeds and under two answer-length limits, and recovers most of the lost mathematics gain. Controls with fixed domain weights show that the gain comes mainly from turning down instruction-following feedback rather than turning up mathematics alone, and that fixed weights close to those DN-MOPD measures perform comparably. Combining specialists therefore requires deciding not only which one teaches, but also how strongly its feedback counts.

cs.LG↗

arXiv2026

Large Language Models for Unit Test Generation: Achievements, Challenges, and Opportunities

Unit tests help detect faults and prevent regressions, but writing them manually is costly. Automated unit test generation involves constructing inputs that exercise the code and oracles that express expected behavior. Large language models (LLMs) can use code and natural-language context to construct tests, but generated tests may fail to compile or execute, contain invalid assertions, or have limited ability to detect faults. We conducted a systematic literature review and mapping study of 178 publications with eligible full-text reports available by 31 July 2026. We organize the literature using a lifecycle-oriented analytical framework that covers test construction and quality assurance. Across the reviewed studies, context enrichment addresses missing project information, while iterative validation and repair use compiler and execution feedback to revise generated tests. These mechanisms help make tests executable, although their contributions to oracle correctness and fault detection vary across tasks and evaluation settings. Meanwhile, benchmark validity, test maintainability, and industrial reliability remain open concerns. Finally, we discuss opportunities in evaluation, context and specification use, hybrid techniques, and agentic workflows.

cs.SE↗

arXiv2026

Artificial intelligences and human scientists exhibit complementary strengths in theory building

We investigate the effectiveness of artificial intelligences (AI)-specifically large language models (LLMs)-relative to human scientists at high-level cognitive tasks in social science such as theory formulation, predictions of novel empirical results, and theory revision in response to new evidence. The research domain was academic discourse regarding gender and race inequality. Our findings, comparing 25 LLMs with 13 senior researchers and 60 doctoral scholars, reveal that the AIs outperformed most humans individually on most of the present tasks, while human theories were more diverse and exhibited greater gains in predictive accuracy from aggregation. AI-generated theories were more extensively elaborated, involving additional theoretical paths and latent variables, and were rated as higher quality than human theories by independent raters blinded to source. However, this theoretical complexity was in part ornamental, in that it was not associated with more accurate predictions about empirical patterns in data; in contrast, human scientists achieved greater predictive efficiency with simpler theories. The AIs were significantly more likely than human scientists to revise their theories to incorporate new evidence; human scientists updated their beliefs in a selective way that is sensitive to prior prediction errors. We speculate that the superior processing capacity of artificial intelligences makes them especially well-suited to tasks requiring grappling with complexity, but that the greater diversity of human ideas is essential to wise crowds and collective creativity.

cs.AI↗

arXiv2026

Conditional Co-Ablation: Recovering Self-Repair Backups in Transformer Circuits

Mechanistic interpretability seeks to explain transformer behavior through circuits: sets of internal components that causally support a behavior. However, self-repair creates a blind spot: ablating a primary component can activate a dormant backup, so a circuit that explains behavior in the intact model can become incomplete under the intervention used to test it. We formulate this gap as conditional circuit completion: given a primary set, identify components that become causally important after its removal. We introduce conditional co-ablation (CoAx), which ranks candidates by growth in ablation effect after primary-set removal. We show that a perfectly dormant backup can be indistinguishable from an irrelevant component to per-unit intact-state scores, whereas its conditional effect change exactly aggregates all interaction orders linking it to the removed set. On GPT-2-small's Indirect Object Identification (IOI) circuit, CoAx recovers the documented backup heads at 0.941 ROC-AUC, versus 0.815 for the strongest intact-state attribution baseline and 0.758 for the matched conditional-energy control. Recovery drops to 0.40 +/- 0.13 AUC for alternative component sets matched in behavioral effect, output displacement, and depth, showing that recovery is specific to the removed circuit. Beyond recovery, the CoAx-selected heads are causally load-bearing: freezing them after primary removal sharply reduces the IOI margin, while adding them to the incomplete circuit reduces incompleteness from 0.75 to 0.21. More broadly, conditional growth aligns with intervention-derived repair in 11/12 held-out instances across 4 mechanism clusters, and CoAx completions outperform matched random completions on all 8 non-GPT-2 models spanning 6 architecture families. Together, causal explanations of self-repairing transformers must account for backup circuitry when primary components fail.

cs.LG↗

arXiv2026

Towards ultra-scaled nanoelectronics using the zipper material system $Bi_2O_2Se/Bi_2SeO_5$

Two-dimensional (2D) materials could overcome the scaling bottleneck of nanoelectronics by enabling atomically thin channels, superior electrostatic control, and reduced short-channel effects. However, progress is limited by the lack of semiconductor–insulator interfaces being simultaneously scalable, stable, and reliable. Conventional 2D interfaces are often low quality or require transferring dissimilar materials, limiting reproducibility and scalability. We show that the zipper heterostructure formed by the high-mobility 2D semiconductor Bi 2 O 2 Se and its native high- κ oxide Bi 2 SeO 5 addresses these challenges, providing an atomically sharp, chemically matched interface with excellent electrostatics and promising scaling potential. We further present the first comprehensive multiscale assessment of Bi 2 O 2 Se/Bi 2 SeO 5 transistors targeting thermal stability, reliability, and scalability by linking atomic-scale structure and defects to device-level behavior across four transistor generations (top-gated, fin, and two gate-all-around architectures). Benchmarking against IRDS-2035 targets indicates that Bi 2 O 2 Se/Bi 2 SeO 5 devices could deliver high drive current with low gate leakage under aggressive scaling, potentially surpassing the targets in the upper-bound region of the sensitivity analysis. Finally, we identify oxygen-related oxide defects as the dominant origin of hysteresis consistent with a beneficial role of encapsulation and oxygen-rich annealing. Together, our findings support the potential of this zipper material system as a technologically-credible and manufacturing-relevant platform for future nanoelectronics.

cond-mat.mtrl-sci↗

arXiv2026

Sparse and weak-measurement certification of graph-edge entanglement in PXP scar wavepackets

An imperfect many-body revival does not by itself certify the entanglement of the returning state. We give a finite-record protocol for graph-edge localizable entanglement along scar wavepackets of a graph-dressed PXP chain. The target cluster state has nonzero energy variance, so neither an exact target eigenstate nor a dark-state embedding is assumed. Fresh binary probes of the undeformed Hamiltonians Pauli terms have a fully separable explanation, even within the dressed blockade sector. Adding local graph-stabilizer probes makes established entanglement witnesses accessible with simultaneous confidence bounds. For a specified square-root instrument, an imposed worst-case disturbance budget fixes the strength that minimizes the equal-allocation Hoeffding sufficient sampling cost. A local commutator bound accounts for finite-duration ancilla pulses while the Hamiltonian remains active. We test the protocol on chains through twenty spins, under perturbed dynamics, and with independent implementations. At the first twenty-spin return, synthetic weak records certify all nineteen graph edges. Generator, two-color, bounded-weight, and full-group benchmarks separate this localizable resource from genuine multipartite certification. The protocol measures recoverable entanglement in a known scar wavepacket, with explicit calibration and limits on its physical and statistical interpretation.

quant-ph↗

arXiv2026

UniPolicy: Unified Objective-Specific Policies for Generative Search Advertising

Search advertising connects user intent with commercial content and plays a critical role in platform monetization. Recent systems typically align pretrained generative models with a single business reward, such as eCPM, or use naive reward fusion for preliminary multi-objective alignment. However, an ideal search advertising system must jointly account for heterogeneous objectives, including relevance, click propensity, and commercial value, to balance user experience and business value while mitigating globally suboptimal performance caused by gradient competition. We propose UniPolicy, an objective-aware multi-policy alignment framework. UniPolicy combines objective-specific prefix tokens, sparse MoE-LoRA routing, and objective-specific residual FFNs to hierarchically decouple parameters within a shared backbone, providing differentiated parameter and policy-expression spaces for different business objectives. It further constructs pairwise preferences from multi-stage behavioral feedback, supplementing the relative preference information in exposed-but-unclicked samples and strengthening the relative advantage of clicked candidates in the generation distribution. At inference, UniPolicy supports parallel, business-customizable multi-policy beam search, flexibly allocating candidate quotas across objectives under a fixed retrieval budget. Large-scale offline experiments show that UniPolicy delivers balanced improvements across multiple metrics while preserving retrieval quality, outperforming single-objective reinforcement learning and naive reward-fusion baselines. In a 7-day online A/B test on a real search advertising system, UniPolicy improves CTR by 0.71%, RPS by 1.58%, and advertising revenue by 1.32%, while maintaining stable serving latency.

cs.CL↗

arXiv2026

Sublattice-selective control of spin reversal in metasurface-coupled Kagome interfaces

Coherent manipulation of photonic interface states requires a control field that matches their internal mode structure. We study this requirement in a two-component Kagome lattice motivated by photon-mediated exchange near a nonlinear nonlocal metasurface. The Dirac spinors show why uniform Raman control cannot couple opposite-spin, same-valley interface modes at leading order, even when their spatial envelopes coincide. A $1:1:-2$ sublattice pattern removes this cancellation and maximizes the projected coupling at fixed root-mean-square amplitude within the diagonal-control class. We test this control scheme through full-zone topology and complete lattice propagation. For a smooth, gapped interface, a $720$-dimensional calculation gives target-mode fidelity $0.999937$ with leakage $6.27\times10^{-5}$ at RMS drive $0.04t$. Trace-preserving dynamics gives the separate survival condition needed for successful conversion. An exploratory three-dimensional lithium-niobate supercell reproduces the complex addressing pattern with $0.373\%$ relative error and provides nonlocal exchange, decay and electro-optic frequency-conversion matrices. The mode-resolved control principle thus gives quantitative electromagnetic design targets; the full spin-dependent device realization still requires further calibration.

quant-ph↗

arXiv2026

Affect-Prototype Guided Fusion for Open-Vocabulary Incomplete Multi-modal Emotion Recognition

Open-vocabulary multimodal emotion recognition (OV-MER) aims to generate open natural-language emotion labels from multimodal affective cues. In real-world scenarios, however, complete and synchronized modal data are difficult to obtain due to limitations of acquisition devices and user privacy constraints. Existing OV-MER methods are largely designed for full-modal inputs, and fail to perform effective feature fusion under modal missing conditions. Meanwhile, current fusion approaches designed for incomplete modalities mainly focus on fixed-label recognition context, and cannot satisfy the demand for fuse emotional cues guided with arbitrary emotion semantics in OV-MER context. To tackle these challenges, this paper proposes an Affect-Prototype-Conditioned Fusion (APCF) framework for incomplete open-vocabulary emotion recognition. As a candidate-free generative framework, APCF extends modal contribution learning to scenarios guided by arbitrary emotional semantics. Specifically, we construct an affect-prototype library to explicitly model multimodal contribution characteristics corresponding to diverse emotions, which provides dynamic constraints for modal fusion under different emotional semantic perspectives. Conditional retrieval and feature aggregation are conducted based on available modal features. The refined fused affective representations are then fed into an LLM decoder to produce open-vocabulary emotion labels. Experiments on the OV-MERD+ and MER-FG datasets demonstrate that APCF substantially outperforms state-of-the-art baselines.

cs.AI↗

arXiv2026

SoftRerank: Hierarchical Soft Fusion with Candidate-Label Reranking for Long-Tailed Micro-Action Recognition

Micro-actions are subtle, low-intensity non-verbal behaviors that provide cues to fine-grained human states, including emotions and intentions. Recognizing them remains difficult because they are brief, contain weak visual changes, and often exhibit similar motion patterns across categories. This paper addresses these challenges with a fine-grained micro-action recognition method that combines full fine-tuning of InternVideo2.5, hierarchical soft fusion, and a lightweight candidate-label reranker. For the long-tailed label distribution in MA-52, we use class-balanced sampling and inverse-frequency reweighting to reduce the effect of frequent classes during training. We fine-tune InternVideo2.5 end to end and attach coarse and group-conditional fine-grained classification heads to the shared video representation, improving the consistency between coarse and fine predictions. For ambiguous samples, the candidate-label reranker uses hard samples and video-label matching to focus on easily confused fine-grained actions. Experiments validate the proposed method, which achieves a 79.99% F1-mean on MA-52 and ranks first in the 3rd Micro-Action Analysis Grand Challenge at ACM Multimedia 2026.

cs.CV↗

arXiv2026

Hi-Singers: A Comprehensive High-Quality Dataset for Expressive Audio-Driven Singing Head Synthesis

State-of-the-art models for audio-driven digital human generation have achieved photo-realistic results in talking-head synthesis. However, extending these models to singing-head synthesis remains challenging due to a significant Domain Gap: singing requires more exaggerated expressions, vivid jaw openings, and precise rhythmic synchronization. Current models, primarily trained on speech datasets, often struggle with "rhythmic drift" and constrained dynamics. To address this, we introduce Hi-Singers, the first large-scale, high-quality, in-the-wild video dataset specifically tailored for singing head synthesis. Hi-Singers undergoes a rigorous automated and manual filtering pipeline, ensuring strict thematic adherence, high-resolution rendering, and stable motion, resulting in 29,608 video segments totaling approximately 170 hours. We further establish a dedicated evaluation benchmark balanced across linguistic and musical styles. Extensive experiments across diverse architectures, including 3D-coefficient and diffusion-based models, demonstrate that Hi-Singers consistently and significantly improves performance across all dimensions. Specifically, it enables models to achieve superior visual realism, enhanced lip-sync consistency, and more precise rhythmic dynamics, effectively bridging the domain gap and setting a new performance standard for the singing synthesis task. The dataset is available at https://huggingface.co/datasets/CharlesZhang-USTC/Hi-Singers

cs.MM↗

arXiv2026

A fully integrated dispersion-managed femtosecond mode-locked laser

Femtosecond lasers underpin applications ranging from material processing to corneal surgery, while their regular pulse trains form optical frequency combs that have revolutionized timekeeping, spectroscopy, and metrology. On-chip optical frequency combs, such as Kerr microcombs, have enabled high-repetition-rate applications in optical communications and microwave photonics. However, integrated chip-scale sources operating at low repetition rates (100 MHz to 1 GHz), crucial for high peak intensities, remain elusive, as existing devices typically operate well beyond 10 GHz. Here, we demonstrate a self-starting, photonic integrated mode-locked laser based on a dispersion-managed architecture that accesses this regime. The laser combines erbium-implanted silicon nitride gain waveguides, integrated chirped Bragg gratings, and a semiconductor saturable absorber mirror to generate optical pulses with repetition rates from 0.5 to 1.2 GHz, pulse durations as short as 300 fs, and mode-locking thresholds down to 27.3 mW. The output forms a passively stable optical frequency comb with a comb-line drift below 1% of the repetition rate, surpassing the stability of commercial fiber lasers by two orders of magnitude. Leveraging this ultra-low threshold, we achieve complete hybrid integration by co-packaging the laser with a telecom-grade 980-nm III-V pump diode chip inside a compact photonic module. The resulting electrical-in/optical-out module delivers turnkey, stable mode-locked pulses, providing a compact, low-power, and vibration-insensitive foundry-compatible platform for field-deployable optical metrology and precision sensing.

physics.optics↗

arXiv2026

MetaStructAtlas: A Grounded 3D Vision-Language Dataset and Benchmark for Functional and Structural Reasoning in Whole-Body PET/CT

The joint interpretation of metabolic function and anatomical structure is essential for clinical diagnosis in whole-body PET/CT. Although recent advances in 3D medical vision-language models have demonstrated remarkable progress, current efforts are limited to regional CT imaging, leaving a critical void in comprehensive whole-body PET/CT analysis. In this work, we introduce MetaStructAtlas, a large-scale dataset for grounded whole-body PET/CT interpretation that synthesizes multimodal imaging with integrated anatomical, metabolic, and semantic annotations. MetaStructAtlas provides 490 co-registered 3D PET and CT volumes with 50,470 organ-level segmentation masks and grounded radiology reports. To facilitate interactive reasoning, we further developed MetaStructVQA, a standardized 3D grounded visual question-answering benchmark containing 100,565 QA pairs. This framework explicitly links diagnostic queries to visual evidence across modalities, encompassing anatomical, morphological, and metabolic characteristics. Finally, we evaluate state-of-the-art 3D medical VLMs on MetaStructVQA, establishing a robust foundation for multimodal representation learning and integrated whole-body reasoning in nuclear medicine.

cs.CV↗

arXiv2026

Low Mach number limit for the Navier--Stokes--Korteweg equations with a stationary force

In this paper, we investigate the low Mach number limit for the three-dimensional compressible Navier--Stokes--Korteweg equations in the whole space under a small stationary external force. We first construct a family of small stationary solutions uniformly with respect to the Mach number $ε$ and prove that both the stationary density fluctuation and the compressible component of the stationary velocity are of order $ε^2$. For ill-prepared non-stationary perturbations around these stationary solutions, we establish the existence and uniqueness of global strong solution by combining uniform high-order energy estimates with a low-frequency Besov estimate and a Kawashima-type compensating functional. The main difficulty is that Korteweg tensor not only changes the elliptic structure of the stationary problem, but also modifies the dispersive mechanism of the acoustic modes. In Korteweg-symmetric variables, the associated spectral projections are uniformly bounded zero-order Fourier multipliers, while the acoustic-capillary phase is wave-like at low frequencies and Schrödinger-like at high frequencies. Since the source terms generated by the stationary coefficients are generally not integrable in time, we decompose the Duhamel source according to its time-integrability and frequency behavior. Dyadic dispersive estimates, high-frequency damping estimates, and maximal regularity for the heat equation yield the global-in-time convergence rate $ε^{\min\{1/r,\,1/2-1/p\}}$ in the mixed Besov norms $L^r(0,\infty;\dot B^s_{p,1})$. As a consequence, Besov embeddings also yield quantitative convergence in the mixed Lebesgue norms $L^r(0,\infty;L^p)$.

math.AP↗

arXiv2026

Lead-Lag Relationships in Financial Markets: A Comparison of Multiple Clustering Algorithms

Lead-lag relationships are widely used in financial time series, and many clustering algorithms based on them have been developed. The traditional DTW-KMedoids algorithm performs well both on the synthetic dataset and the real financial dataset. However, there are still several limitations to these algorithms: low efficiency caused by high time complexity, poor mathematical properties from DTW distance, the clustering effect is sensitive to the number of clusters. To solve the problems above and improve the performance, this paper introduces three clustering algorithms: MiniRocket-KMeans, KShape, Ensemble algorithm (a combination of KShape and DTW-KMedoids) and compares their performance on synthetic and real stock datasets with DTW-KMedoids algorithm under the same trade strategy. In addition, this paper also finds the best number of clusters by maximizing the silhouette coefficient in each clustering algorithm to improve the stability of the experiment results. Our main conclusions are as follows: MiniRocket-KMeans performs best under the lead strategy, achieving a Sharpe ratio of 0.866 with a maximum drawdown controlled at -63.9\%; the ensemble algorithm exhibits excellent stability; the robustness is significantly improved after finding the best number of clusters; the p-values of the hypothesis test on the Sharpe ratio of all strategies are 0.0, verifying the statistical validity of the lead-lag trading strategy. Finally, future improvement directions such as customized lead-lag matrices and optimized ensemble voting mechanisms are proposed.

q-fin.ST↗

arXiv2026

UR$^{2}$-MLLM: Uncertainty-aware Revisit Reasoning in Multimodal Large Language Models for Radiology Report Generation

Radiologists generate diagnostic reports through iterative and selective revisiting of suspicious regions to refine their interpretations. Recent multimodal large language models (MLLMs) for radiology report generation (RRG) have shifted from text-only reasoning toward a ``Thinking-with-Images'' paradigm, incorporating visual evidence into the reasoning process. However, existing methods provide static visual evidence without a dynamic revisit mechanism during reasoning, neglecting how radiologists re-examine uncertain observations. To this end, we propose an Uncertainty-aware Revisit Reasoning MLLM (UR$^{2}$-MLLM) framework that dynamically revisits uncertain regions during reasoning for RRG. UR$^{2}$-MLLM is first equipped with uncertainty perception by training on an uncertainty-aware dataset. We then construct a multimodal reasoning trajectory dataset together with a detect-and-copy mechanism, which guides when and where to revisit. Finally, a visual grounding reward refines this behavior through reinforcement learning, aligning the revisited regions with corresponding anatomical structures. Experiments on MIMIC-CXR and IU-Xray show that UR$^{2}$-MLLM achieves state-of-the-art performance, highlighting the value of uncertainty-aware visual revisit reasoning for reliable and clinically aligned report generation.

cs.CV↗

arXiv2026

Context Is Not Authority: Structured Runtime Governance for Financial Market Agents

Financial agents can turn correct context into an unauthorized effect: a customer-facing commitment, trade, or deployed policy. We present SAGE-Fin, a finance-specific authority-handoff contract that makes the proposed effect, not merely its text, the object of runtime control. SAGE-Fin compiles proposals into typed, adapter-bound candidates; records missing or stale institutional obligations as coverage debt; contracts authority under current market, account, policy, and dialogue state; and requires an exact-artifact receipt whose nominal type matches the consuming response, execution, or policy adapter. Evidence and workflow progress cannot substitute for effect authority, and prior authorization is rechecked after state changes. Across an authored 616-case catalog, five deterministic specifications yield 3,080 outputs; a label-isolated harness obtains 616/616 binary reference-prototype parity, including 3/3 named response-gate fixtures, while 22 tests cover selected paths. These results establish executable conformance, not independent safety accuracy. Separately, SAGE-Fin's response gate processed real customer-facing production requests at a confidential digital-asset platform. An operational team independent of the implementation team reached a strongly positive post-deployment conclusion on practical usefulness and workflow fit, and end-user feedback was also strongly positive. Disclosure permits only the review's independence, stakeholder classes, assessed dimensions, and directional conclusion, so this is qualitative field corroboration rather than an aggregate effect estimate. Three distinct de-identified predecessor failures, with independently confirmed 0/3 interception, ground repeated-emission drift, stale account evidence, and missing escalation state without estimating prevalence or treatment effect.

cs.AI↗

arXiv2026

On the Graphical Rules for Recovering the Average Treatment Effect Under Selection Bias

Selection bias is a major obstacle toward valid causal inference in epidemiology. Over the past decade, several graphical rules based on causal diagrams have been proposed as sufficient identification conditions for addressing selection bias and recovering causal effects. However, these simple graphical rules are typically coupled with specific identification strategies and estimators. In this article, we show two important cases of selection bias that fall outside the scope of these existing simple rules and their estimators: one case where selection is a descendant of a collider of the treatment and the outcome, and the other case where selection is affected by a mediator. To address selection bias and recover the marginal average treatment effect in these two cases, we propose an alternative set of graphical rules and construct identification formulas using g-computation and inverse probability weighting (IPW) based on single-world intervention graphs (SWIGs), leveraging external information from individuals outside the selected sample. We conduct simulation studies to verify the performance of the estimators when the traditional crude selected-sample analysis (i.e., complete-case analysis) yields erroneous conclusions that contradict the truth. We emphasize that identifying causal effects in the presence of selection bias depends critically on both the target estimand and the availability of external information.

stat.ME↗