arXiv subjects

Yunfei Zhang

Publications and source records attributed to Yunfei Zhang.

At least 19 recordsLinked to original sources

arXiv2026

RobustSGPO: Search-Space Control for Agent Harness Evolution

Semantic-gradient-based prompt optimization (SGPO) improves agent harnesses using execution feedback, but its local update rule leaves the choice of edit scope and operation unresolved. We introduce RobustSGPO, which specifies the requested edit, constructs and checks the patch, and continues search from either the incumbent or retained snapshots. We evaluate permission scheduling, cumulative controls, and task-family transfer in the AgentX brainstorming workflow using 120 tasks, 95 runs, and 7,350 candidate attempts. Periodic $1\to2\to3$ scheduling exceeds fixed maximum permission by 0.28 test-score points. RobustSGPO increases completion on 30 held-out tasks from 60.0% to 80.0% and improves test quality from 3.77 to 4.14 under a 20-million-token budget. Category retention reduces source-task degradation after a shift, whereas random retention reaches a higher destination endpoint. Search-space control benefits quality through executable edits and alternative starting points, with measurable retention overhead.

cs.AI↗

arXiv2026

LongRCA Bench: Diagnosing Responsible Roles and Root Causes in Long-Horizon Agent Failures

When a long-horizon agent execution fails, outcome-level evaluation reveals the unsuccessful result but not where the decisive error entered the trajectory. Developers must then inspect the full execution to identify the responsible role and localize the earliest decisive root-cause step. Existing failure-attribution benchmarks largely focus on shorter traces, leaving diagnosis across hundreds of recorded steps underexplored. We introduce LongRCA Bench, comprising 1,140 failed trajectories across five domains without injected errors. It provides independently scored human labels for the responsible role and earliest decisive root-cause step. The median trajectory contains 145 steps, and the strongest baseline reaches only 13.2% exact root-step accuracy. We further present Root-Cause Trajectory Attribution (RCTA), a training-free method that retrieves candidate error steps from segment summaries and traces them to available earlier handoff instructions. Using the same backbone, benchmark instances, and scoring protocol, RCTA reaches 51.1% responsible-role accuracy and 24.1% exact root-step accuracy. These results highlight the need to evaluate responsible-role attribution and exact root-step localization as separate targets in long-trajectory failure diagnosis.

cs.AI↗

arXiv2026

InstancePin: Instance-Addressable Layout-to-Image Diffusion via Coordinate Pinning

Layout-to-image diffusion models have achieved impressive semantic controllability by conditioning generation on category-level segmentation maps. However, such category-aligned control is not necessarily instance-addressable: multiple nearby objects from the same category are often treated as a shared semantic region, leading to ambiguous boundaries, averaged appearances, and feature confusion among instances. This limitation is particularly evident in urban scene synthesis, where small and crowded pedestrians or vehicles require fine-grained instance separation while preserving global scene consistency. In this paper, we propose InstancePin, an instance-addressable layout-to-image diffusion framework that pins each object instance with an explicit coordinate anchor. Instead of directly injecting instance masks into the pretrained backbone, InstancePin introduces an independent instance-aware adapter to preserve the category-level generation prior while learning instance-specific spatial control. For each instance, its center coordinate is encoded with Fourier features and projected into a coordinate token, which serves as a spatial anchor queried by latent image features through coordinate pinning attention. To make these anchors spatially meaningful, we further supervise the coordinate attention maps with instance regions, encouraging each coordinate token to activate its corresponding object area. Finally, an instance-mask guided fusion module routes pretrained backbone features to non-instance regions and adapter features to instance regions, enabling local instance refinement without sacrificing global semantic fidelity. Extensive experiments on Cityscapes demonstrate that InstancePin mitigates instance entanglement in dense layouts and improves both image fidelity and semantic consistency.

cs.CV↗

arXiv2026

MECA: A Mechanism-Centered Agent for Constructing Well-Specified and Valuable Mathematical Conjectures

Automatically constructing well-specified and valuable mathematical conjectures remains a central challenge in AI-assisted mathematical discovery. Many existing open problems and conjectures are often too broad, underspecified, or difficult to connect to plausible proof or refutation strategies. We view a mathematical mechanism as a structure or reasoning principle that connects the assumptions of a candidate problem to its target conclusion, such as an inequality, invariant, decomposition, or reduction to an intermediate claim. We present MECA (MEchanism-centered Conjecture Agent), a multi-agent framework that constructs conjectures by jointly developing candidate statements and their supporting mechanisms. Explorer agents propose mechanisms, test how they apply, and revise the candidate conjecture accordingly, while critic agents assess their mathematical validity and research value. Their feedback guides changes to the assumptions, scope, and conclusion. Through this process, MECA transforms broad research directions into precise conjectures with substantive mathematical support while retaining a clearly identified unresolved core. We evaluate MECA in two complementary settings. First, we compare it with a generate-and-revise baseline on reconstructing preselected target-paper conclusions from target-conditioned but article-blind source materials. Second, we construct 100 semi-open problems from literature-derived seeds and existing open problems and evaluate them through independent proof and refutation attempts by automated provers. Our results indicate that mechanism-centered refinement produces well-specified and research-worthy conjectures that remain challenging for current automated provers.

cs.AI↗

arXiv2026

Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces

The emergence of Large Language Models (LLMs) has illuminated the potential for a general-purpose user simulator. However, existing benchmarks remain constrained to isolated scenarios, narrow action spaces, or synthetic data, failing to capture the holistic nature of authentic human behavior. To bridge this gap, we introduce OmniBehavior, the first user simulation benchmark constructed entirely from real-world data, integrating long-horizon, cross-scenario, and heterogeneous behavioral patterns into a unified framework. Based on this benchmark, we first provide empirical evidence that previous datasets with isolated scenarios suffer from tunnel vision, whereas real-world decision-making relies on long-term, cross-scenario causal chains. Extensive evaluations of state-of-the-art LLMs reveal that current models struggle to accurately simulate these complex behaviors, with performance plateauing even as context windows expand. Crucially, a systematic comparison between simulated and authentic behaviors uncovers a fundamental structural bias: LLMs tend to converge toward a positive average person, exhibiting hyper-activity, persona homogenization, and a utopian bias. This results in the loss of individual differences and long-tail behaviors, highlighting critical directions for future high-fidelity simulation research.

cs.CL↗

arXiv2026

CAM-Bench: A Benchmark for Computational and Applied Mathematics in Lean

Formal theorem-proving benchmarks enable mechanically verifiable evaluation of mathematical reasoning in large language models. However, existing benchmarks mainly focus on Olympiad-style problems and algebraic domains, leaving computational and applied mathematics underrepresented. We introduce CAM-Bench, a Lean 4 theorem-proving benchmark of 1,000 Lean proof targets in computational and applied mathematics, with coverage spanning optimization, numerical linear algebra, and numerical analysis. These problems are adapted from textbook exercises and often depend on locally introduced definitions, notation, algorithms, and elementary results. To construct CAM-Bench, we develop a dependency-recovery pipeline that reconstructs the local textbook context needed to state each problem faithfully. It then normalizes each problem into a standalone informal theorem and translates it into a Lean target. We validate the resulting formal problems through Lean compilation and semantic review, checking both formal correctness and semantic alignment with the original exercises. For each problem, we release the raw exercise, recovered context, normalized informal theorem, and final Lean target. CAM-Bench complements existing formal mathematics benchmarks by targeting applied mathematics problems that rely on textbook concepts and elementary theorems, many of which are not directly available as standard Mathlib4 lemmas. We evaluate widely used large language models and formalization agents on CAM-Bench, and analyze common failure modes in tracking local assumptions, applying elementary results, decomposing proofs, and maintaining long-horizon control in Lean.

cs.AI↗

arXiv2026

Unlocking Exploration in RLVR: Uncertainty-aware Advantage Shaping for Deeper Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) has shown significant promise for enhancing the reasoning capabilities of large language models (LLMs). However, prevailing algorithms like GRPO broadcast a uniform advantage signal across all tokens in a sequence. This coarse-grained approach overlooks the pivotal role of uncertain, high-stakes decisions during reasoning, leading to inefficient exploration and the well-documented problem of entropy collapse. To address this, we introduce UnCertainty-aware Advantage Shaping (UCAS), a model-free method that refines credit assignment by leveraging the model's internal uncertainty signals. UCAS operates in two stages: it first modulates the response-level advantage using a logit-space self-confidence proxy, and then applies an asymmetric token-level penalty based on raw logit certainty. This dual mechanism encourages exploration of high-uncertainty paths that yield correct answers while penalizing overconfident yet erroneous reasoning, effectively balancing the exploration-exploitation trade-off. Extensive experiments on five mathematical reasoning benchmarks show that UCAS significantly outperforms strong RLVR baselines across multiple model scales, including 1.5B and 7B. Our analysis confirms that UCAS not only achieves higher rewards but also promotes greater reasoning diversity and successfully mitigates entropy collapse. Code is available at https://github.com/xvolcano02/UCAS.

cs.AI↗

arXiv2025

ChromouVQA: Benchmarking Vision-Language Models under Chromatic Camouflaged Images

Vision-Language Models (VLMs) have advanced multimodal understanding, yet still struggle when targets are embedded in cluttered backgrounds requiring figure-ground segregation. To address this, we introduce ChromouVQA, a large-scale, multi-task benchmark based on Ishihara-style chromatic camouflaged images. We extend classic dot plates with multiple fill geometries and vary chromatic separation, density, size, occlusion, and rotation, recording full metadata for reproducibility. The benchmark covers nine vision-question-answering tasks, including recognition, counting, comparison, and spatial reasoning. Evaluations of humans and VLMs reveal large gaps, especially under subtle chromatic contrast or disruptive geometric fills. We also propose a model-agnostic contrastive recipe aligning silhouettes with their camouflaged renderings, improving recovery of global shapes. ChromouVQA provides a compact, controlled benchmark for reproducible evaluation and extension. Code and dataset are available at https://github.com/Chromou-VQA-Benchmark/Chromou-VQA.

cs.CV↗

arXiv2025

Study on impact mechanism and precursor information induced by high intensity mining

With heightened mining intensity, the incidence of coal bursts is escalating, necessitating advanced understanding and prediction techniques. This research delves into the intricacies of coal burst mechanisms, proposing a novel theoretical model for the release of coal mass energy founded on the tenets of stress superposition. A significant revelation is that the energy culminating in a coal burst is an amalgamation of intrinsic coal strain energy and perturbations from mining activities. Field investigations scrutinize the microseismic parameters across a spectrum of mining velocities, discerning potential failure regions and precursor hallmarks in high-intensity mining environments. Notably, microseismic energy, in such contexts, experiences an augmentation of approximately 2000 J. Numerical simulations executed via 3DEC elucidate stress distribution patterns and failure modalities of adjacent rock structures in relation to mining velocities. The simulations underscore that an uptick in mining speed diminishes the buffer to high-pressure abutments, intensifying inherent pressures. For mitigation, it's advocated that high-intensity mining advances be capped at 11 m/d. Merging theoretical analysis, experimental data, field assessments, and computational simulations, this study proffers a holistic insight into coal burst dynamics, underscoring its value in refining monitoring and early warning protocols in the domain.

physics.geo-ph↗

arXiv2025

Discrete Prior-based Temporal-coherent Content Prediction for Blind Face Video Restoration

Blind face video restoration aims to restore high-fidelity details from videos subjected to complex and unknown degradations. This task poses a significant challenge of managing temporal heterogeneity while at the same time maintaining stable face attributes. In this paper, we introduce a Discrete Prior-based Temporal-Coherent content prediction transformer to address the challenge, and our model is referred to as DP-TempCoh. Specifically, we incorporate a spatial-temporal-aware content prediction module to synthesize high-quality content from discrete visual priors, conditioned on degraded video tokens. To further enhance the temporal coherence of the predicted content, a motion statistics modulation module is designed to adjust the content, based on discrete motion priors in terms of cross-frame mean and variance. As a result, the statistics of the predicted content can match with that of real videos over time. By performing extensive experiments, we verify the effectiveness of the design elements and demonstrate the superior performance of our DP-TempCoh in both synthetically and naturally degraded video restoration.

cs.CV↗

arXiv2025

Enhancing Neural Adaptive Wireless Video Streaming via Lower-Layer Information Exposure and Online Tuning

Deep reinforcement learning (DRL) demonstrates its promising potential in the realm of adaptive video streaming and has recently received increasing attention. However, existing DRL-based methods for adaptive video streaming use only application (APP) layer information, adopt heuristic training methods, and train generalized neural networks with pre-collected data. This paper aims to boost the quality of experience (QoE) of adaptive wireless video streaming by using lower-layer information, deriving a rigorous training method, and adopting online tuning with real-time data. First, we formulate a more comprehensive and accurate adaptive wireless video streaming problem as an infinite stage discounted Markov decision process (MDP) problem by additionally incorporating past and lower-layer information, allowing a flexible tradeoff between QoE and costs for obtaining system information and solving the problem. In the offline scenario (only with pre-collected data), we propose an enhanced asynchronous advantage actor-critic (eA3C) method by jointly optimizing the parameters of parameterized policy and value function. Specifically, we build an eA3C network consisting of a policy network and a value network that can utilize cross-layer, past, and current information and jointly train the eA3C network using pre-collected samples. In the online scenario (with additional real-time data), we propose two continual learning-based online tuning methods for designing better policies for a specific user with different QoE and training time tradeoffs. Finally, experimental results show that the proposed offline policy can improve the QoE by 6.8~14.4% compared to the state-of-arts in the offline scenario, and the proposed online policies can further achieve 6~28% gains in QoE over the proposed offline policy in the online scenario.

cs.MM↗

arXiv2024

A rotational ellipsoid model for solid Earth tide with high precision

Solid Earth tide represents the response of solid Earth to the lunar (solar) gravitational force. The yielding solid Earth due to the force has been thought to be a prolate ellipsoid since the time of Lord Kelvin, yet the ellipsoid's geometry such as major semi-axis's length, minor semi-axis's length, and flattening remains unresolved. Additionally, the tidal displacement of reference point is conventionally resolved through a combination of expanded potential equations and given Earth model. Here we present a geometric model in which both the ellipsoid's geometry and the tidal displacement of reference point can be resolved through a rotating ellipse with respect to the Moon (Sun). We test the geometric model using 23-year gravity data from 22 superconducting gravimeter (SG) stations and compare it with the current model recommended by the IERS (International Earth Rotation System) conventions (2010), the average Root Mean Square (RMS) deviation of the gravity change yielded by the geometric model against observation is 6.47 μGal (equivalent to 2.07 cm), while that yielded by the current model is 30.77 μGal (equivalent to 9.85 cm). The geometric model will greatly contribute to many application fields such as geodesy, geophysics, astronomy, and oceanography.

physics.geo-ph↗

arXiv2024

Temporal Correlation Meets Embedding: Towards a 2nd Generation of JDE-based Real-Time Multi-Object Tracking

Joint Detection and Embedding (JDE) trackers have demonstrated excellent performance in Multi-Object Tracking (MOT) tasks by incorporating the extraction of appearance features as auxiliary tasks through embedding Re-Identification task (ReID) into the detector, achieving a balance between inference speed and tracking performance. However, solving the competition between the detector and the feature extractor has always been a challenge. Meanwhile, the issue of directly embedding the ReID task into MOT has remained unresolved. The lack of high discriminability in appearance features results in their limited utility. In this paper, a new learning approach using cross-correlation to capture temporal information of objects is proposed. The feature extraction network is no longer trained solely on appearance features from each frame but learns richer motion features by utilizing feature heatmaps from consecutive frames, which addresses the challenge of inter-class feature similarity. Furthermore, our learning approach is applied to a more lightweight feature extraction network, and treat the feature matching scores as strong cues rather than auxiliary cues, with an appropriate weight calculation to reflect the compatibility between our obtained features and the MOT task. Our tracker, named TCBTrack, achieves state-of-the-art performance on multiple public benchmarks, i.e., MOT17, MOT20, and DanceTrack datasets. Specifically, on the DanceTrack test set, we achieve 56.8 HOTA, 58.1 IDF1 and 92.5 MOTA, making it the best online tracker capable of achieving real-time performance. Comparative evaluations with other trackers prove that our tracker achieves the best balance between speed, robustness and accuracy. Code is available at https://github.com/yfzhang1214/TCBTrack.

cs.CV↗

arXiv2023

ReliCD: A Reliable Cognitive Diagnosis Framework with Confidence Awareness

During the past few decades, cognitive diagnostics modeling has attracted increasing attention in computational education communities, which is capable of quantifying the learning status and knowledge mastery levels of students. Indeed, the recent advances in neural networks have greatly enhanced the performance of traditional cognitive diagnosis models through learning the deep representations of students and exercises. Nevertheless, existing approaches often suffer from the issue of overconfidence in predicting students' mastery levels, which is primarily caused by the unavoidable noise and sparsity in realistic student-exercise interaction data, severely hindering the educational application of diagnostic feedback. To address this, in this paper, we propose a novel Reliable Cognitive Diagnosis(ReliCD) framework, which can quantify the confidence of the diagnosis feedback and is flexible for different cognitive diagnostic functions. Specifically, we first propose a Bayesian method to explicitly estimate the state uncertainty of different knowledge concepts for students, which enables the confidence quantification of diagnostic feedback. In particular, to account for potential differences, we suggest modeling individual prior distributions for the latent variables of different ability concepts using a pre-trained model. Additionally, we introduce a logical hypothesis for ranking confidence levels. Along this line, we design a novel calibration loss to optimize the confidence parameters by modeling the process of student performance prediction. Finally, extensive experiments on four real-world datasets clearly demonstrate the effectiveness of our ReliCD framework.

cs.CY↗

arXiv2023

Data-driven Spatio-Temporal Scaling of Travel Times for AMoD Simulations

With the widespread adoption of mobility-on-demand (MoD) services and the advancements in autonomous vehicle (AV) technology, the research interest into the AVs based MoD (AMoD) services has grown immensely. Often agent-based simulation frameworks are used to study the AMoD services using the trip data of current Taxi or MoD services. For reliable results of AMoD simulations, a realistic city network and travel times play a crucial part. However, many times the researchers do not have access to the actual network state corresponding to the trip data used for AMoD simulations reducing the reliability of results. Therefore, this paper introduces a spatio-temporal optimization strategy for scaling the link-level network travel times using the simulated trip data without additional data sources on the network state. The method is tested on the widely used New York City (NYC) Taxi data and shows that the travel times produced using the scaled network are very close to the recorded travel times in the original data. Additionally, the paper studies the performance differences of AMoD simulation when the scaled network is used. The results indicate that realistic travel times can significantly impact AMoD simulation outcomes.

eess.SY↗

arXiv2023

Autonomous Vehicles as a Sensor: Simulating Data Collection Process

Urban traffic state estimation is pivotal in furnishing precise and reliable insights into traffic flow characteristics, thereby enabling efficient traffic management. Traditional traffic estimation methodologies have predominantly hinged on labor-intensive and costly techniques such as loop detectors and floating car data. Nevertheless, the relentless progression in autonomous driving technology has catalyzed an increasing interest in capitalizing on the extensive potential of on-board sensor data, giving rise to a novel concept known as "Autonomous Vehicles as a Sensor" (AVaaS). This paper innovatively refines the AVaaS concept by simulating the data collection process. We take real-world sensor attributes into account and employ more accurate estimation techniques based on the on-board sensor data. Such data can facilitate the estimation of high-resolution, link-level traffic states and, more extensively, online cluster- and network-level traffic states. We substantiate the viability of the AVaaS concept through a case study conducted using a real-world traffic simulation in Ingolstadt, Germany. The results attest to the ability of AVaaS in estimating both microscopic (link-level) and macroscopic (cluster- and network-level) traffic states, thereby highlighting the immense potential of the AVaaS concept in effecting precise and reliable traffic state estimation and also further applications.

eess.SP↗

arXiv2023

Effects of Dynamic and Stochastic Travel Times on the Operation of Mobility-on-Demand Services

Mobility-on-Demand (MoD) services have been an active research topic in recent years. Many studies focused on developing control algorithms to supply efficient services. To cope with a large search space to solve the underlying vehicle routing problem, studies usually apply hard time-constraints on pick-up and drop-off while considering static network travel times to reduce computational time. As travel times in real street networks are dynamic and stochastic, assigned routes considered feasible by the control algorithm in one time step might become infeasible in the next. Since once assigned and confirmed, customers should still be part of the solution, damage control is necessary to counteract this effect. In this study, a detailed simulation framework for MoD services is coupled with a microscopic traffic simulation to create dynamic and stochastic travel times, and tested in a case study for Munich, Germany. Results showed that the combination of inaccurate travel time estimation and damage control strategies for infeasible routes deteriorates the performance of MoD services -- hailing and pooling -- significantly. Moreover, customers suffer from unreliable pick-up time and travel time estimations. Allowing re-assignments of initial vehicle schedules according to updated system states helps to restore system efficiency and reliability, but only to a minor extent.

eess.SY↗

arXiv2022

FleetPy: A Modular Open-Source Simulation Tool for Mobility On-Demand Services

The market share of mobility on-demand (MoD) services strongly increased in recent years and is expected to rise even higher once vehicle automation is fully available. These services might reduce space consumption in cities as fewer parking spaces are required if private vehicle trips are replaced. If rides are shared additionally, occupancy related traffic efficiency is increased. Simulations help to identify the actual impact of MoD on a traffic system, evaluate new control algorithms for improved service efficiency and develop guidelines for regulatory measures. This paper presents the open-source agent-based simulation framework FleetPy. FleetPy (written in the programming language "Python") is explicitly developed to model MoD services in a high level of detail. It specially focuses on the modeling of interactions of users with operators while its flexibility allows the integration and embedding of multiple operators in the overall transportation system. Its modular structure ensures the transferabillity of previously developed elements and the selection of an appropriate level of modeling detail. This paper compares existing simulation frameworks for MoD services and highlights exclusive features of FleetPy. The upper level simulation flows are presented, followed by required input data for the simulation and the output data FleetPy produces. Additionally, the modules within FleetPy and high-level descriptions of current implementations are provided. Finally, an example showcase for Manhattan, NYC provides insights into the impacts of different modules for simulation flow, fleet optimization, traveler behavior and network representation.

cs.MA↗