arXiv subjects

Xiaoyu Zhang

Publications and source records attributed to Xiaoyu Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Scaling Articulated Rationales for MLLM-based Recommendation

We presented SARA, an industrial framework that transforms sparse articulated user rationales into scalable recommendation signals. Its data engine curates questionnaire responses into SARA-HQ, providing explicit preference supervision for aligning SARA-7B through SFT and Quality-Refining DPO. This alignment extends rationale generation from $86{,}564$ questionnaire-covered authors to the full $10$M-author space. SARA-Ranker translates the generated positive and negative rationales into features for user--author interaction modeling and negative-feedback history modeling, connecting articulated reasons to production ranking. Evaluation on unseen authors demonstrates that SARA-7B generates more specific, relevant, and grounded rationales than the evaluated general-purpose MLLMs. On top of a strong industrial ranking baseline with multimodal features, separate online A/B tests show that positive-rationale integration increases watch time by $0.99\%$, while negative-rationale integration reduces Hate feedback by $8.16\%$. Daily refresh and more than $30$ days of production deployment further demonstrate the operational feasibility of the approach. These findings establish articulated rationales as a useful complement to behavioral and content signals, and demonstrate a practical role for MLLMs in scaling sparse human explanations into preference information that improves industrial recommendation.

cs.IR↗

arXiv2026

Notrix: Understanding Machine Learning Solutions Across Computational Notebooks at Scale

Computational notebooks make problem-solving visible, but typically only one notebook at a time. Meanwhile, in data science platforms like Kaggle, one competition can accumulate hundreds of notebooks. Effective collection-level analysis requires characterizing recurring solution patterns across all notebooks, as well as isolating specific notebooks for closer examination and learning. However, standard notebooks provide no common basis for this. Their workflows are nonlinear, cells declare no intent, and identical code can serve different ends, leaving hundreds of notebooks as separate documents. In this paper, we present Notrix, an interactive visual analytics tool for profiling hundreds of notebooks as one collection. Inspired by a formative study (N = 11), Notrix classifies every cell into one of thirteen machine learning (ML) stages, turning each notebook into a stage sequence, and clusters those sequences by structure rather than by code. To keep the representation constant as the scope narrows from the whole collection to a single cell, Notrix features three coordinated views---Workflow, Structural Matrix, and Detail---that appear at all four levels of granularity. In a within-subject study (N = 17) using two Kaggle collections of over 400 notebooks each, we observed participants answered questions about all notebooks more accurately with Notrix (median 88% vs. 50%) while opening 80% fewer notebooks per minute. Notably, four of the fourteen answered it without opening a single notebook (interaction logs, N = 14). Participants also reported significantly lower mental demand, temporal demand, and stress with Notrix (Holm-Bonferroni adjusted).

cs.HC↗

arXiv2026

PIVOT: Physics-Grounded Verification for AI-Generated Audio-Video Detection

As generative models continue to advance, AI-generated content (AIGC) is becoming increasingly realistic, weakening the artifact cues commonly exploited by existing detectors. Nevertheless, faithfully reproducing the physical behavior of real-world events remains challenging for current generators. We therefore explore detecting AIGC by assessing whether the depicted event satisfies measurable constraints derived from physical laws. We introduce PIVOT, a physics-grounded AIGC detector, instantiated here for audio-video clips, that estimates physical quantities from video and audio, selects physical laws relevant to each clip, and verifies their measurable constraints. Beyond a real/fake decision, PIVOT returns supporting evidence that records the verification outcome, relevant time window, and supporting quantities for each applicable law. Although instantiated and evaluated here on audio-video data, the framework can, in principle, extend to other AIGC modalities whenever the physical quantities required for verification can be estimated reliably. We also introduce PhysForensics-Bench, comprising paired real and generated audio-video clips from nine event-centric scene families and two recent audio-video generators. On PhysForensics-Bench, PIVOT achieves 70.30% accuracy and 64.29% F1 score on Real+Seedance, and 72.16% accuracy and 65.82% F1 on Real+VEO. In comparison, direct inspection with Gemini 3.1 Pro obtains 53.96% accuracy and 60.09% F1 on Real+Seedance, and 57.22% accuracy and 63.44% F1 on Real+Veo. These results demonstrate the practical promise of physical-consistency verification as a structured and inspectable source of evidence that complements artifact-based AIGC detection.

cs.CV↗

arXiv2026

Input-to-state stability of chemical reaction networks with application to molecular computation

Input-to-state stability (ISS) provides a useful tool for analyzing the robustness of time-varying chemical reaction networks (CRNs). This paper investigates the ISS property for CRNs in two cases: one is for weakly reversible CRNs with nonzero deficiency and multiple linkage classes, which is beyond the limit of existing results requesting a weakly reversible CRN to be zero-deficiency and single-linkage-class; the other is for non-weakly reversible CRNs by leveraging the notion of linear conjugacy. These results, on the one hand enrich the studies of the ISS property of broader classes of CRNs, on the other hand serve for designing CRN-based molecular computing systems. The latter application suggests the ISS theory can justify the parallel mechanism of biomolecular computations. We use some CRNs of practical relevance to demonstrate our results, including the p53 signaling and dimerization network, the p21-activated kinase 1 network, etc.

math.DS↗

arXiv2026

KnowVis: Knowledge-Centric Visual Summarization for Video Lectures

Video lectures are valuable educational resources, but their dense and lengthy formats often overwhelm novice learners. This difficulty stems from a fundamental pedagogical mismatch: while videos deliver transient information linearly, human learning requires constructing interconnected cognitive networks, a task that induces severe cognitive overload for novice learners lacking prior domain knowledge. Existing video summarization methods fail to resolve this mismatch, as they primarily produce text-heavy, linear condensations that still demand high cognitive effort. To bridge this gap, we propose KnowVis, a framework that transforms linear video lectures into pedagogically grounded visual narratives. KnowVis first extracts a detailed concept map from multimodal video content to identify important and challenging threshold concepts, then constructs structured knowledge units, and finally synthesizes engaging visual summaries. Alongside the framework, we introduce a curated dataset of 125 educational videos across 10 academic disciplines, paired with 1,079 generated visual summaries. Extensive automated evaluations and a human study demonstrate that, compared to state-of-the-art baselines, KnowVis generates more accurate and clear visuals that successfully reduce cognitive load and significantly improve student learning effectiveness and knowledge retention.

cs.CV↗

arXiv2026

Reduced-Rank Network Autoregression with Grouped Edge Effects

We propose the Edge-Grouped Reduced-Rank Network Autoregressive model (Edge-Grouped RRNAR) for multivariate time series observed over a network. The model allows transmission effects to vary across prespecified and economically interpretable groups of edges, while using a low-rank structure to capture cross-variable dynamics. This structure separates where network transmission occurs from which variables transmit and respond. We develop a topology-aware and block-specific scaled gradient descent algorithm with a convex low-rank initialization, and establish its local linear convergence and non-asymptotic estimation rates. We further derive asymptotic normality for the estimated transition matrix and normalized grouped network coefficients, enabling Wald tests for network relevance and edge-group homogeneity. Simulations demonstrate the finite-sample performance of the proposed estimation and inference procedures. An application to quarterly U.S. industry data linked by the production network reveals distinct predictive transmission intensities across economically defined edge groups, identifies interpretable real-activity and price-cost channels, and improves forecasting relative to standard network and matrix autoregressive models.

stat.ME↗

arXiv2026

Robustness Analysis of Agentic AI to Inconsistent and Incomplete Tool Responses

Tool-using agents increasingly rely on external tools to complete multi-step tasks, but tool returns can fail in different ways and require different recovery actions. Existing robustness studies often use uncertainty-based measures to detect when an agent becomes unreliable. These measures can reveal that something has gone wrong, but they do not directly identify the type of tool failure or the appropriate response. We address this limitation by analyzing tool failures at the moment a return enters the agent context. Our approach combines two complementary signals. The first compares the likelihood of the returned content under the tool schema and under the full trajectory prefix. The second measures the agent's probability distribution over its legal next actions. We evaluate the approach by injecting incomplete and inconsistent returns into a retail customer-service benchmark. The results show that likelihood-based signals clearly capture incomplete returns and some direct inconsistencies, while action-based signals reveal how strongly a failure changes the next decision. Some failures that are weak under likelihood signals can still redirect the agent toward state-changing actions. These findings show that tool failures can be recognized at the return boundary, but reliable diagnosis requires combining multiple signals.

cs.AI↗

arXiv2026

Repair or Resample? Rethinking Failure Debugging in LLM Multi-Agent Systems

As large language model (LLM)-based multi-agent systems (MASs) are increasingly applied to long-horizon complex tasks, their reliability has emerged as the core bottleneck hindering their real-world deployment. Existing MAS debugging and repair methods typically rely on rerunning and resampling the entire execution trajectory. However, a fundamental question remains to be answered: do these methods causally repair MAS failures or merely stochastically repair by leveraging the randomness of LLM sampling? To evaluate the effectiveness of MAS repair methods, we introduce SymTrace, a controlled evaluation framework that records the MAS execution trajectory and establishes intervention anchors. During replay, it effectively reconstructs the execution before the anchor using recorded logs and only regenerates the downstream trajectory, thereby enabling the reliable reproduction of MAS failures. We further construct the dataset SymFail, comprising 536 human-annotated failure trajectories with graph-linked locations, categories, and trace evidence. Based on these foundations, we conduct a large-scale empirical study across three mainstream MAS frameworks. Our findings reveal that existing unguided rerun methods are highly unreliable, exhibiting low failure reproduction and repair rates (only 67.97% and 6.90%, respectively). Building upon these findings, we further explore the effectiveness of a symptom-driven intervention method, which successfully repairs 20.15% of the failed cases (a 191.89% improvement to state-of-the-art repair methods). This study aims to provide actionable insights for MAS debugging and repair research, paving the way for the robust deployment of multi-agent systems.

cs.AI↗

arXiv2026

A-CPES: A Reference Framework for Agentic AI in Cyber-Physical Energy Systems

Energy system operation contains a loop of work that automation has never taken over: posing the optimization problem the current cycle should solve, disposing of infeasibility, sequencing a solution into interlocked switching orders, assembling evidence no single model holds, negotiating adjustable capacity with many parties, and settling experience into practice. Licensed dispatchers carry all of it in person, and the rising share of variable renewable generation is making that loop turn faster than their number can grow. Agentic AI supplies the abilities it requires, but enters as the outer loop of control: it calls SCED and the other decision models rather than being called by them. We propose A-CPES, three nested rings, an authorization and accountability frame around an agentic control outer loop around a six-layer CPES core. We argue the loop is indivisible, tune where and how tightly it may close, state eight structural failure modes as falsifiable predictions, and specify six governance modules that rebuild the authorization frame until it covers the loop, before the loop starts turning.

cs.AI↗

arXiv2026

Real-Time Density--Source Mapping for Open Electron--Photon Systems

Real-time quantum-electrodynamical density-functional theory (QEDFT) for an open device requires local electron and field histories to determine their sources. For occupation-dependent interactions and a gapped finite-band photon continuum, we prove a causal one-to-one map on the forward image over a nonzero time interval. Three source-reconstruction benchmarks test the electronic limit, the photon-continuum limit, and a coupled joint inverse against a closed-form continuum endpoint. This provides the density--source foundation for real-time open-system QEDFT.

physics.chem-ph↗

arXiv2026

FlashAccel: Leveraging High-Bandwidth Flash (HBF) for High-Throughput LLM Inference

Large language model (LLM) inference is increasingly limited by the capacity of High-Bandwidth Memory (HBM) in GPUs, as model weights and KV cache grow rapidly. High-Bandwidth Flash (HBF) provides higher capacity than HBM while offering comparable bandwidth, making it a promising substrate for capacity-constrained LLM inference. However, its inherently high access latency, low bandwidth utilization, and lack of support for heterogeneous resource management make it difficult to integrate HBF into GPUs for LLM inference. We present FlashAccel, a co-designed system that enables efficient LLM inference using HBF. FlashAccel integrates HBF into HBM-based GPUs, providing architectural support to mitigate access latency. It improves bandwidth utilization through specialized data layouts for both model weights and KV cache, and introduces an HBF-aware storage management layer together with a programming model to organize persistent data in HBF and coordinate heterogeneous memory resources at the system level. Experimental results demonstrate that integrating six HBF stacks into the GPU enables FlashAccel to deliver an average improvement of 2.49$\times$ and 1.93$\times$ in throughput per GPU and energy efficiency over the HBM-only GPU under a 100ms latency constraint, respectively.

cs.AR↗

arXiv2026

Pretrained Optimization Model for Zero-Shot Black Box Optimization

Zero-shot optimization involves optimizing a target task that was not seen during training, aiming to provide the optimal solution without or with minimal adjustments to the optimizer. It is crucial to ensure reliable and robust performance in various applications. Current optimizers often struggle with zero-shot optimization and require intricate hyperparameter tuning to adapt to new tasks. To address this, we propose a Pretrained Optimization Model (POM) that leverages knowledge gained from optimizing diverse tasks, offering efficient solutions to zero-shot optimization through direct application or fine-tuning with few-shot samples. Evaluation on the BBOB benchmark and two robot control tasks demonstrates that POM outperforms state-of-the-art black-box optimization methods, especially for high-dimensional tasks. Fine-tuning POM with a small number of samples and budget yields significant performance improvements. Moreover, POM demonstrates robust generalization across diverse task distributions, dimensions, population sizes, and optimization horizons. For code implementation, see https://github.com/ninja-wm/POM/.

cs.NE↗

arXiv2026

Finite-Temperature Spin-Adapted ROKS and TDDFT

Finite-temperature conditions constitute a central regime of interest in quantum chemistry. Nonetheless, a consistent incorporation of finite-temperature effects into density functional theory (DFT) for both ground and excited states, while rigorously preserving the spin-symmetry associated with the $\hat{S}^2$ operator, has not yet been achieved. In this work, we develop a finite-temperature extension of restricted open-shell Kohn-Sham (ROKS) theory and spin-adapted time-dependent density functional theory (TDDFT), providing unified frameworks for the description of ground and excited states, respectively. For finite-temperature ROKS, we construct a canonical ensemble by using integer high-spin ROKS components, where each component $I$ has the same spin number $|SS \rangle$ and its weight $w_I$ follows the Boltzmann distribution. For finite-temperature spin-adapted TDDFT, every integer component $I$ supplies an ordinary zero-temperature spin-adapted TDDFT matrix pair $(\Mmat_I,\Nmat_I)$. These matrices are arranged in one unified spatial-orbital order and then averaged. In the zero-temperature single-component limit, the theory reduces to ordinary high-spin ROKS and the corresponding spin-adapted TDDFT. As a numerical application, we apply the theory to a diradicaloid and use the calculated excitations to interpret the thermally activated absorption observed in variable-temperature UV/Vis spectroscopy.

physics.chem-ph↗

arXiv2026

Where Agent Frameworks Fall Short: Examining Functional Challenges and Usability Concerns

Large language model (LLM) agents are increasingly built on agent frameworks that provide reusable abstractions for workflow orchestration, state management, tool integration, and execution control. However, the quality of this infrastructure layer remains insufficiently understood, particularly its functionality challenges and usability concerns, as existing studies have mainly examined traditional deep learning (DL) frameworks or model-level agent failures. Therefore, we conduct an empirical study of 5,669 bug reports and 809 feature requests from five mainstream agent frameworks: AutoGen, CrewAI, LangChain, LangGraph, and MetaGPT. We construct a four-dimensional taxonomy covering 22 root causes, seven symptoms, 11 motivations, and six requirements, and map them to the five-stage agent lifecycle. Across the four RQs, results show an execution centered quality pattern shaped by semantic interface boundaries. Reported bugs mainly manifest as Incorrect Functionality (76.00%) and involve more API, configuration, parsing, and serialization related causes than DL framework bugs, while their associations remain sparse and stage specific. Feature requests mainly target Feature Enhancement (49.07%) and reveal structured needs for Orchestration Expressiveness, Development Delivery, Model Adaptation, and Tool Ecosystem. These findings call for quality assurance beyond crash based and tensor level testing, with emphasis on API sequences, structured LLM outputs, serialization boundaries, execution traces, and execution centered maintenance, offering empirical guidance for reliable and usable agent framework infrastructure.

cs.SE↗

arXiv2026

CodeChemist: Test-Time Scaling for Low-Resource Code Generation via Functional Knowledge Transfer

Code Large Language Models (CodeLLMs) have been widely adopted for Natural Language to Programming Language code generation, powering applications with large user bases. Their performance, however, varies sharply across programming languages (PLs) and is particularly suboptimal for low-resource PLs due to data scarcity, limiting their overall usability. In this work, we introduce CodeChemist, a simple yet effective, training-free test-time scaling framework that transfers the model's functional knowledge from high-resource to low-resource PLs via synthesized test cases, without relying on external models. Specifically, CodeChemist first applies multi-temperature hedged sampling to generate a pool of candidate solutions in the low-resource PL and synthesizes a set of test inputs. It then estimates uncertainty: when uncertainty is low, it selects the output via in-language majority voting; otherwise, it constructs cross-lingual I/O test oracles by executing high-resource reference programs and selects the candidate with the highest pass rate. Extensive experiments demonstrate that CodeChemist significantly outperforms existing test-time scaling methods, improving code generation for both low-resource PLs (e.g., Lua) and complex-syntax PLs (e.g., C++, Java) without retraining.

cs.SE↗

arXiv2026

QuerySplat: Decoupling Geometry and Appearance Representations in 3DGS Prediction

While feed-forward 3D Gaussian Splatting (3DGS) enables efficient 3D reconstruction, achieving high-fidelity rendering remains challenging. Existing pixel-aligned approaches suffer from spatial inflexibility and massive structural redundancy, whereas query-based methods lack 3D priors and entangle geometry with appearance, yielding blurry, pose-dependent results. To overcome these deficiencies, we propose \textbf{QuerySplat}, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling. Specifically, we design a dual-branch query-based decoder: the geometry branch leverages a pretrained Vision Geometric Model for spatial understanding, which intrinsically endows QuerySplat with pose-free modeling capabilities, while the appearance branch recovers high-frequency details through a dedicated pathway separated from geometric attribute regression. Extensive experiments demonstrate that QuerySplat mitigates the blurry rendering issues of early query-based models and consistently outperforms pixel-aligned approaches in rendering fidelity. On the challenging DL3DV benchmark, it achieves state-of-the-art novel view synthesis performance, with average PSNR gains of 2.30 dB and 1.04 dB over the best pose-free and pose-required baselines, respectively. Project Page: https://inspatio.github.io/querysplat.

cs.CV↗

arXiv2026

End-to-End Differentiable Learning of a Single Functional for DFT and Linear-Response TDDFT

Density functional theory (DFT) and linear-response time-dependent density functional theory (LR-TDDFT) rely on an exchange-correlation (xc) approximation that provides not only energy but also its functional derivatives that enter the self-consistent potential and the response kernel. Here, we present an end-to-end differentiable workflow to optimize a single deep-learned energy functional using targets from both Kohn-Sham DFT and adiabatic LR-TDDFT. To enable this training in a computationally efficient and differentiable manner, we developed a JAX-based two-component quantum chemistry package (IQC), in which the learned functional provides a self-consistent potential and linear-response kernel via automatic differentiation. This construction permits gradient-based optimization through both the self-consistent-field (SCF) fixed-point equations and the Casida eigenvalue problem. We learn an exchange-correlation functional on excitation energies and ground-state properties (noncovalent interactions, thermochemistry, bond dissociation, ionization potentials, electron affinities, isomerization energies, and reaction barriers) while incorporating one-electron self-interaction cancelation as penalty terms, and we assess its possible transfer to molecular test cases.

physics.chem-ph↗

arXiv2026

Translating Spin-Adapted RPA to Spin-Adapted TDDFT

Linear-response TDDFT based on open-shell references may produce spin-mixed excited states, particularly when the response space is not spin complete. Spin-adapted RPA was developed using tensor equation-of-motion and applying the Wigner-Eckart theorem with tensor decoupling. Casting the RPA Fock matrix and kernels as energy derivatives gives a TDDFT extension. To restore spin-component degeneracy and eliminate multi-counting of spin correlation in pure exchange-correlation functionals, we use hybrids combining HF exchange with spin-unpolarized pure XC parts. Given that the mapping from spin-adapted RPA to spin-adapted TDDFT is \textit{ad hoc}, we introduce a first-principles framework for evaluating the expectation value $\langle \hat{S}^2 \rangle$ of excited states in order to rigorously prove their spin purification. In addition, we derive explicit working equations for the computation of the single-particle density matrix of excited states, thereby enabling a natural orbital analysis of their electronic structure. Benchmarks, diradicals/triradicals, Cr$_2$ dissociation, and phenol O-H conical intersections demonstrate the method.

physics.chem-ph↗