Searcharxiv⌕ Search

arXiv subjects

Zihao Chen

Publications and source records attributed to Zihao Chen.

At least 19 recordsLinked to original sources
arXiv2026

MBABench-2.0: Evaluating AI Agents on Constructing Financial Spreadsheets Beyond Simple Verifiability

Spreadsheets are the lingua franca of modern finance, driving valuation, forecast, audit, and many other core workflows of the industry. As LLM agents advance to automate knowledge work, spreadsheet construction is a natural and pertinent testbed. However, most existing spreadsheet benchmarks evaluate agents on fill-in-the-blank style tasks, where an agent performs minor edits in otherwise complete spreadsheets. This protocol fails to capture the most central and time-consuming part of the spreadsheet workflow: the modeling decisions, format design, and sheet layout regular knowledge workers need to toil through to build a workbook from scratch. We introduce MBABench-2.0, a benchmark of 101 original spreadsheet construction tasks, the hardest of which is estimated to take 33 hours for a modeler with 3+ years of experience. Unlike coding or math, where a unit test or a final answer can verify an agent's output, the quality of a financial spreadsheet that concerns readability, maintainability, or auditability, is not so simply verifiable. To holistically evaluate spreadsheet construction, we introduce 129 criteria spanning 12 categories that operationalize quality beyond the accuracy of the spreadsheet. Evaluating over 20 agents across 8 models and 4 harness types, we find that frontier agents rarely construct spreadsheets that are both accurate and meet our quality criteria: the top agent achieves a pass rate of only 16% on the benchmark. The benchmark demonstrates a substantial gap in spreadsheet building capability of current agents.

cs.SE↗
arXiv2026

Noise-shaped Sampling and Annealing for Scalable On-the-edge Training of Binary and Ternary AI Models

The training loss of binary and ternary neural networks, expanded as a multilinear polynomial over the discrete weights, is structurally equivalent to the energy of a hierarchical high-order Ising model. We exploit this equivalence to apply Ising solvers to discrete-weight network training, while accounting for a fundamental difference in objectives: an Ising solver seeks the ground state of a fixed energy function, whereas training seeks solutions that generalize to unseen data. Our key contribution is $ΣΔ$-Sampler, an optimizer that reconciles these objectives by embedding a stochastic sampling-and-annealing step as a noisy quantizer inside a first-order $ΣΔ$ noise-shaping loop. The loop pushes noise from weight quantization, high-order frustration, and mini-batch sampling out of the low-frequency adaptation band, so that discrete weight updates follow the smooth descent trajectories embedded in the rugged loss landscape. We validate the framework on a planted quadratic unconstrained binary optimization (QUBO) benchmark with a known minimizer; on CIFAR-10, where a $13$M-parameter ternary model comes within $1\%$ of full-precision accuracy; on ImageNet-1k, where a ternary ResNet-18 comes within $0.4\%$ Top-5 accuracy of a ternary baseline; on class-incremental continual learning, where prior-task accuracy is preserved; and on MatMul-free language models up to $2.7$B parameters, where it outperforms a straight-through-estimator baseline under matched continued pre-training. Compared with standard quantization-aware training, $ΣΔ$-Sampler reduces the training memory footprint by over $50\%$, delivers an order-of-magnitude speed-up, and reduces weight-update memory traffic by four orders of magnitude, making it attractive for on-device training and fine-tuning of large binary and ternary models.

cs.LG↗
arXiv2026

REGEN: Replay-recycling for Expert-to-Generalist distillation with Offline Reinforcement Learning

Large-scale online reinforcement learning (RL) is the predominant means of eliciting advanced abilities including long-term reasoning and agentic tool use in large language models (LLMs). However, continuing to scale it across vast task domains of interest remains challenging in both computational infrastructure and cost, especially when considering RL as merely a one-off learning stage. Recently, a widely used technique for distilling knowledge across various domains and training stages, multi-teacher on-policy distillation (MOPD), helps to decouple the RL stage, saving costs, while maintaining generality across vast domains. Nonetheless, similar to online RL, MOPD requires coupled inference and backward passes, which continues to limit its scalability and computational efficiency. To address these challenges, we propose REGEN: Replay-recycling for Expert-to-Generalist Distillation with Offline RL. Instead of distilling from multiple teacher models, REGEN trains a generalist by simply recycling the replay memory -- the free by-product of the teachers' specialized RL training -- and employing offline RL algorithms. REGEN completely decouples the rollout sampling from the backward training process and thus greatly reduces the training cost. Across mathematical reasoning, code generation, and instruction following, REGEN matches the accuracy of MOPD at substantially lower cost. It potentially turns online RL into a data synthesis process instead of a one-off learning stage, and can be extended to large-scale post-training without requiring heavy computational load. Code is available at https://github.com/yunjie-sysu/REGEN.

cs.LG↗
arXiv2026

Physical AI Smart Spaces: A Large-Scale Benchmark for Multi-Camera 3D Perception in Smart Spaces

Physical AI Smart Spaces is, to the best of our knowledge, the first benchmark to simultaneously provide large-scale, multi-class, and multi-camera 3D perception data for indoor smart spaces. It contains over 280 hours of synchronized 1080p footage captured by nearly 1,800 cameras in warehouses, hospitals, retail venues, and similar settings, together with automatic annotations for multi-camera identities, 2D bounding boxes, 3D bounding boxes, camera calibration, and depth where available. The benchmark spans Isaac Sim synthetic generation, Cosmos Transfer appearance augmentation, and real-world Sim2Real evaluation. For the real-world target, we include two warehouse deployments with time-synchronized streams, automatic VGGT-based calibration, and a 3D labeling interface that projects world-frame 3D boxes into each view for cross-camera verification. We describe the dataset scope, annotation and calibration schema, generation workflow, benchmark protocols, and official evaluation system, which standardizes submission format, and leaderboard reporting. A central contribution is a 3D instantiation of Higher Order Tracking Accuracy (HOTA), extending the usual 2D box-based tracking evaluation to 3D locations and 3D boxes. We further report empirical baselines from the AI City Challenge leaderboards, showing how methods evolve from person-only 3D location tracking to multi-class 3D box tracking under realistic smart-space constraints. The release is available at https://huggingface.co/datasets/nvidia/PhysicalAI-SmartSpaces.

cs.CV↗
arXiv2026

SERA: Scale-Equalized Rollout Allocation for Maximum Likelihood Reinforcement Learning

Maximum Likelihood Reinforcement Learning (MaxRL) targets prompt-wise log-success and has shown strong performance on reasoning tasks. Under finite rollout budgets, however, the estimator used by MaxRL attenuates each prompt's likelihood gradient by a factor that depends on its success probability and rollout count. Under uniform rollout allocation, the common rollout count fails to compensate for success-dependent attenuation, leaving low-success prompts more strongly attenuated and distorting their relative contributions to the expected aggregate gradient. We introduce SERA (Scale-Equalized Rollout Allocation), which redistributes a fixed rollout budget to approximately equalize these finite-rollout scaling factors. Building on our theoretical analysis of how finite rollouts distort prompt-wise likelihood gradients, we formulate the allocation as a fixed-budget max--min problem, derive a waterline solution to its continuous relaxation, and introduce a multiplicity correction to remove the additional prompt weighting induced by heterogeneous rollout counts. Experiments show stronger alignment with exact likelihood gradients in a controlled ImageNet setting and improved multi-sample solution coverage over MaxRL on maze navigation and mathematical reasoning under matched training rollout budgets.

cs.LG↗
arXiv2026

AgentBug-Smith: Automatically Reproducing Real-World Harness Bugs in Agentic Systems

Agent harness bugs exhibit unique characteristics and remain challenging for state-of-the-art software agents to repair. Progress in this area is further hindered by existing benchmarks, which contain only a small and fixed number of executable harness bugs while requiring hundreds of human hours to construct. This work presents AgentBug-Smith, an automated harness bug reproduction approach that continuously discovers and reproduces real-world harness bugs from open-source agentic systems. Across different backbone LLMs, AgentBug-Smith consistently outperforms existing bug reproduction techniques designed for general software, achieving 10.67% - 27.56% higher success rates of reproducing harness bugs. By applying AgentBug-Smith to open-source agentic systems in the wild, we construct Live-Harness-Bench, a live and extensible benchmark that currently contains 200 reproducible harness bugs. We further demonstrate the utility of Live-Harness-Bench through two downstream applications. First, we use Live-Harness-Bench as the evaluation benchmark to systematically evaluate state-of-the-art software agents, revealing their limited capabilities in repairing real-world harness bugs. Second, we use Live-Harness-Bench as a knowledge base of real-world harness bug fixes, from which reusable repair skills can be distilled to improve existing software agents, increasing their harness-bug repair rates by 6.32%. Together, AgentBug-Smith and Live-Harness-Bench establish a scalable foundation for continuously evaluating and improving software agents on harness bug repair, turning real-world agent failures into executable evaluation instances and reusable knowledge for harness improvement, thus contributing to the ultimate goal of recursively self-improving agents.

cs.SE↗
arXiv2026

When Do Surrogate Metrics Work? A Finite-Sample Comparison Under Realistic Failure Modes

Online experiments must often be evaluated before long-term outcomes mature. Under rolling enrollment, these outcomes are observed only for early enrollees, while short-term surrogates are available for everyone. We compare seven estimators across eleven data-generating processes, spanning partial mediation, drift, outcome sparsity, and enrollment-time labeling, with up to $R = 2,000$ replications over more than 500 method-by-scenario cells. We find a sharp robustness-efficiency tradeoff: the surrogate index delivers large efficiency gains when surrogacy holds but its coverage collapses under violations, while PPI-family methods stay asymptotically valid under random labeling at smaller gains. We give the finite-sample variance of PPI++ in the all-units parameterization for a fixed predictor, a joint asymptotic distribution for the two estimators under cross-fitting, and a Hausman-type estimator-disagreement diagnostic, then quantify the detection-damage gap: in the partial-mediation design, where we locate both edges, a band of violations destroys surrogate-index coverage yet is too small to detect on most datasets. On the 64,000-customer Hillstrom experiment the diagnostic rarely flags a violation that biases the surrogate index, and a Cauchy-kernel hybrid of the two estimators inherits 10.7% relative bias; on the 14-million-user Criteo experiment the violation is detected, and subsampling traces detection turning on with scale as damage persists. PPI++ has limits: at a rare-conversion $n = 30,000$ Criteo subsample its empirical coverage is 85.0%. We recommend prespecifying PPI++ with the exact variance as the primary analysis under random labeling and adequate labeled outcome counts, reading the diagnostic as a warning, not a certificate, and treating the surrogate index as a sensitivity analysis.

stat.ME↗
arXiv2026

Spectral side channels of wavelength-division multiplexer in quantum key distribution under laser damage

In the transmitter of a quantum key distribution (QKD) system, a wavelength-division multiplexer (WDM) is typically used to combine quantum and synchronization signals and is directly connected to the quantum channel. As a result, it becomes the first optical component exposed to laser-injection attacks. Therefore, understanding the behavior of WDMs under such attacks is essential for assessing the practical security of QKD systems. In this work, we systematically investigate the characteristics of WDMs under high-power laser illumination. Our experimental results show that certain WDM samples exhibit pronounced changes in their spectral features once the injected laser power surpasses a specific threshold. Taking the Trojan-horse attack as an illustrative example, we further perform a theoretical analysis of the resulting spectral side channel and show that it can reduce the maximum secure transmission distance to below $66.9\%$ of its original value. By combining experimental observations with theoretical modeling, this study advances the understanding of the influence of WDMs on the practical security of QKD systems.

quant-ph↗
arXiv2026

CoMerge: Conflict-Driven Preference Optimization for Multi-Task Model Merging

Model merging provides an efficient paradigm for constructing multi-task large language models (LLMs) without full model retraining, yet it remains challenged by parameter interference. While existing methods aim to preserve the capabilities of individual expert models and mitigate interference, they generally do not directly learn from the potentially degraded behaviors exposed by naive merging. In this paper, we propose a conflict-driven preference optimization framework for model merging (CoMerge), which reformulates model merging as a preference optimization problem. The approach utilizes a self-supervised, conflict-driven strategy that leverages the defects of naive merging methods (e.g., task arithmetic) as hard negative samples to construct preference pairs without external annotations. By applying preference optimization to refine lightweight, tensor-wise merging coefficients, CoMerge enables the model to mitigate parameter-space conflicts while preserving task-specific capabilities. Extensive experiments show that CoMerge achieves an average normalized performance of 0.9968 on MergeBench, outperforming all evaluated data-free and data-driven model-merging baselines. Furthermore, on Llama-3.1-8B-Instruct, CoMerge yields marked improvements on conflict-sensitive tasks such as instruction following and safety, while remaining highly competitive with full-parameter fine-tuning despite optimizing only 1,445 scalar coefficients.

cs.AI↗
arXiv2026

AutoDRI: Bridging the Semantic Gap for Automated Design Rule Integration in CP-SAT-Based Cell Synthesis under Multi-Patterning

Design-rule integration (DRI) remains a major bottleneck for scalable (Constraint Programming with SAT) CP-SAT-based standard cell synthesis and rapid technology enablement at advanced nodes. It still depends heavily on manual effort and domain expertise. Moreover, existing low-level rule encodings are not expressive enough for emerging constraints such as cut-based rules under multi-patterning technology. This paper presents \textbf{AutoDRI}, a multi-agent framework for automated design-rule integration in standard cell synthesis. AutoDRI combines a geometric semantic library, a standardized conflict-set encoding, a constructive multicolor-cut modeling method, and a feedback-driven multi-agent flow to bridge the semantic gap between natural-language design rules and executable CP-SAT constraints. In the reported experiments, AutoDRI achieves near-perfect rule-integration correctness across 41 cell benchmarks under 10+ complex rules, including colored cut-mask spacing rules, reaching 33/33 correct integrations with Gemini-3-pro and 32/33 with GPT-5.4, while maintaining runtime comparable to manual hard-coding and passing KLayout DRC and Cadence LVS.

eess.SY↗
arXiv2026

IF:CARGO: LLM-Based Semantic Compilation for Al-Native Rule Programming Games

This case study presents IF: CARGO, an experimental puzzle game that uses a large language model as a semantic compiler rather than an autonomous game-playing agent. Players author IF/THEN rules in natural language, which the model translates into a constrained command schema for deterministic validation and execution by the game engine. This architecture creates a playable loop of expression, execution, observation, and revision, framing AI interaction as semantic debugging. A mixed-methods playtest with 24 participants across eight levels examined player attempts, thinking time, perceived controllability, adjustability, and interpretations of the AI's role. Results suggest that players generally understood the model as a translation intermediary and could revise their strategies through feedback, while periodic commands, multi-robot coordination, and rule-priority mechanics created greater cognitive and diagnostic demands. The study proposes a practical pattern for AI-native gameplay: constrain natural-language input, preserve player authorship, and ensure deterministic execution.

cs.HC↗
arXiv2026

PinEqualizer: Full Funnel Content Exploration and Debiasing System at Pinterest

In this paper, we propose a new solution for addressing the content cold-start problem in industry-scale search and recommender systems. Compared to prior approaches, we have made the following new contributions: 1) our solution spans the entire multi-stage funnel and generalizes well for both search and recommendation surfaces, 2) our solution reduces bias favoring existing content, allowing more accurate model prediction across content types and reducing short-term tradeoffs associated with high volumes of explicit content exploration, 3) our solution is evaluated with a scalable measurement framework that enables fast short-term experimentation while validating long-term impact. We have iteratively built and successfully deployed this new system at Pinterest in the past two years and observed significant improvements in fresh content exploration, overall user engagement, and content ecosystem health.

cs.IR↗
arXiv2026

PhyAgentOS: A Self-Evolving Operating System for Embodied Agents with Decoupled Cognitive Planning and Physical Execution

Vision-language-action models, world models, and agentic planners each advance physical intelligence, yet their composition lacks a common execution abstraction, shared state, semantic verification, and persistent experience across heterogeneous embodiments. We present PhyAgentOS, a runtime foundation delivering scheduling, verification, memory, benchmarking, and safety as system-level services. Its Session-Centered Runtime treats a session, not an action, as the minimum unit of scheduling, compatibility preflight, supervised execution, evidence collection, and acceptance. To decouple cognition from physical execution, the cognition-physics boundary is a file system: the State-as-a-File protocol materializes cross-layer state as Markdown with YAML, yielding inspectable, versionable records without code dependencies between Agent and Runtime layers. These views form a unified cognitive state space aligning intent, capabilities, environment, execution, and experience. The SessionVerifier distinguishes execution termination from semantic task completion via evidence-grounded verdicts of success, failure, or replan. Verified outcomes are consolidated through epistemic memory into reusable knowledge and corrective lessons, closing a trial-and-error loop without retraining. Benchmarking reuses the deployment session and verification path, so results trace to real execution. Layered safety constrains both policy-driven and agent-driven execution: preflight, action bridges, SafetyGuard, heartbeat monitoring, and target-local constraints. Validation is progressive: games test cognitive planning, simulation adds dynamics and control, real robots add hardware noise, with the cognitive layer held constant. PhyAgentOS is benchmarked on Optimus-67, StarDojo, and DST-Dojo, validated on 19+ simulated and physical embodiments, and gains on LIBERO, Calvin, and RoboCasa365 across multiple VLA models.

cs.RO↗
arXiv2026

YingMusic-Singer: Controllable Singing Voice Synthesis with Flexible Lyric Manipulation and Annotation-free Melody Guidance

Regenerating singing voices with altered lyrics while preserving melody consistency remains challenging, as existing methods either offer limited controllability or require laborious manual alignment. We propose YingMusic-Singer, a fully diffusion-based model enabling melody-controllable singing voice synthesis with flexible lyric manipulation. The model takes three inputs: an optional timbre reference, a melody-providing singing clip, and modified lyrics, without manual alignment. Trained with curriculum learning and Group Relative Policy Optimization, YingMusic-Singer achieves stronger melody preservation and lyric adherence than Vevo2, the most comparable baseline supporting melody control without manual alignment. We also introduce LyricEditBench, the first benchmark for melody-preserving lyric modification evaluation. The code, weights, benchmark, and demos are publicly available at https://github.com/ASLP-lab/YingMusic-Singer-Plus.

eess.AS↗
arXiv2026

Dataset Construction for Training LLM to Learn Analog Circuit Knowledge

This paper constructs a textual dataset for training large language models (LLMs) to learn analog circuit knowledge and customizes LLM training techniques. For dataset construction, high-quality textbooks are collected and decomposed into fine-grained learning nodes, which are then used to construct structured question-thinking-solution-answer (QTSA) quadruples using a multi-agent framework to capture both final answers and thought processes. The resulting dataset consists of 7.26M tokens of unlabeled data for continual pre-training (CPT) and 112.65M tokens of labeled data for supervised fine-tuning (SFT). We customize the training techniques including initial model selection, training paradigms, regularization techniques, and practical implementation references. Instruct models are identified as suitable training initialization points, an SFT-centric training paradigm is established (finding that CPT provides marginal benefits compared with SFT due to imbalanced data distribution), and SFT with KL divergence regularization can achieve a 2.71 percentage-point improvement over SFT alone. A practical training implementation method is provided for resource-constrained scenarios. Experiments demonstrate that the dataset and training techniques enhance LLMs' analog circuit knowledge. The trained 32B instruct model achieves 84.59% accuracy on the AMSBench-TQA benchmark, showing a 15.67 percentage-point improvement over the initial model. The trained model also shows capability in the operational amplifier design task based on the Atelier framework.

cs.AR↗
arXiv2026

UniVoice: A Unified Model for Speech and Singing Voice Generation

Text-to-speech (TTS) and singing voice synthesis (SVS) both aim to generate human vocal audio from symbolic inputs, but they impose different requirements on the generation process. Speech generation relies on flexible, language-driven prosody, whereas singing generation requires explicit melody control and accurate rhythmic alignment. This mismatch makes it challenging to train a single model that can generate both natural speech and controllable singing, since melody-related conditions should strongly constrain singing but should not restrict speech prosody. We present UniVoice, a unified speech and singing voice generation framework based on conditional flow matching. Instead of using a single undifferentiated conditioning representation, UniVoice factorizes the condition into content, melody, and timbre, which are encoded by modality-appropriate encoders and consumed by a shared Diffusion Transformer (DiT) backbone. For singing, the melody condition is represented by MIDI note sequences; for speech, it is replaced with a learned null melody token, allowing the model to infer prosody from linguistic and acoustic context. This design preserves explicit melody control for singing while avoiding the need to impose melody constraints on speech. We further analyze the null melody token as an approximation to melody marginalization in the conditional flow. Trained on 30k hours of speech and 35k hours of singing data, UniVoice achieves a speech PER of 5.26\%, comparable to dedicated TTS systems such as F5-TTS (5.21\%) and CosyVoice3 (5.30\%). On singing generation, UniVoice achieves a PER of 16.22\%, outperforming the unified baseline Vevo1.5 (24.72\%).

cs.SD↗
arXiv2026

rt2gtfs: A scalable framework for correcting public transport timetables using real-time data for accessibility analysis

Travel time is a fundamental component of accessibility measurement, yet most accessibility analyses rely on static timetable data that assume public transport services operate exactly as scheduled. Such representations overlook the substantial variability in travel times arising from operational conditions and service disruptions. In this paper, we present rt2gtfs, an open-source Python package for reconstructing empirical public transport timetables from high-frequency vehicle location data. The package provides a configurable and scalable workflow for collecting GTFS-Realtime vehicle position feeds from the UK Bus Open Data Service (BODS), matching observed vehicle locations to scheduled GTFS trips and stops, inferring stop-level arrival and departure times, and exporting corrected GTFS format timetable bundles. Using national-scale real-time bus data feeds from BODS, we demonstrate how rt2gtfs can be used to generate observed timetables suitable for routing and origin-destination travel time calculation, as well as accessibility analysis. By packaging the framework as reusable software, this work supports more reproducible and realistic accessibility analysis and provides a practical tool for researchers and practitioners seeking to incorporate observed public transport performance into transport planning.

cs.CY↗
arXiv2026

A Unifying View of Anchoring via Operator-Side Tikhonov Regularization

Anchored fixed point and monotone equation methods, including Halpern iteration, extra anchored gradient, and their relatives, add a vanishing pull toward a reference point to obtain last-iterate guarantees. Existing anchored variants often achieve sharp last-iterate guarantees, but from the update-level perspective the placement of the anchor can be algorithm-specific and conceptually opaque. We show that anchoring admits a single operator-side construction: regularize the operator queried by the base method with a vanishing Tikhonov term, then run the unmodified base method. Applied to the Picard iteration, this recipe reproduces the Halpern iteration; applied to the forward step, extragradient (EG), and past extragradient (PEG, also known as Popov's method), it yields three variants whose anchor placements inherit the base method's query pattern. The forward-step instantiation gives a new residual convergence guarantee, while the EG and PEG instantiations give new regularized variants. The four analyses share a residual recurrence, recovering the $O(1/k)$ Halpern residual-norm convergence rate, giving $O(1/\sqrt{k})$ for the regularized forward step, and giving $O(1/k)$ for the regularized EG and PEG variants in the unconstrained monotone Lipschitz setting.

math.OC↗