arXiv subjects

Yingqing Guo

Publications and source records attributed to Yingqing Guo.

6 recordsLinked to original sources

arXiv2026

LC-GRPO: Bridging Train-Inference Gap for Flow-Based GRPO with Langevin Correction

Flow-based generative models are typically sampled by solving a deterministic ordinary differential equation (ODE), whereas online reinforcement learning requires stochastic rollouts for policy exploration and optimization. Existing GRPO methods for flow models therefore replace the inference-time ODE with a stochastic differential equation (SDE) during training. Although the ODE and SDE share the same marginal distributions in continuous time, their finite-step discretizations can differ substantially. In particular, SDE rollouts often become blurry as the exploration noise increases, creating a mismatch between the samples used for reinforcement learning and those generated by the test-time ODE sampler. We introduce LC-GRPO, a flow-based GRPO framework with Langevin correction. Each rollout transition first takes an inference-aligned ODE Euler step and then applies a stochastic Langevin correction targeting the marginal distribution at the resulting timestep. The required score is recovered directly from the flow velocity, requiring no additional score model, while the resulting transition remains an isotropic Gaussian with a tractable likelihood for policy optimization. We theoretically show that, under suitable conditions, one Langevin correction step reduces the Wasserstein error of an imperfect ODE Euler step. At a matched randomness level, we further show that the proposed transition can be more accurate than the standard Euler--Maruyama discretization of the reverse SDE. Experiments on SD3.5-Medium, FLUX.1-Dev, and HunyuanVideo demonstrate that LC-GRPO consistently improves reward optimization across text-to-image and text-to-video tasks, preserves generation quality, and substantially narrows the gap between stochastic training rollouts and deterministic test-time ODE inference.

cs.LG↗

arXiv2025

Joint System Modeling Approach for Fault Simulation of Start-er/Generator and Gas Generator in All-Electric APU

This paper presents a joint system modeling approach for fault simulation of all-electric auxiliary power unit (APU), integrating starter/generator turn-to-turn short circuit (TTSC) faults with gas generator gas-path faults.To address challenges in electromechanical coupling, simulation precision and computational efficiency balance, we propose a multi-rate continuous-discrete hybrid simulation architecture. This architecture treats the starter/generator as a continuous system with variable step size in Simulink, while modeling the gas generator as a discrete system with fixed step size in a dynamic-link library (DLL) environment. For the starter/generator fault modeling, a multi-loop approach is deployed to accurately simulate TTSC faults. For the gas generator, we develop an improved GasTurb-DLL modeling method (IGDM) that enhances uncertainty modeling, state-space representation, and tool chain compatibility. Finally, the proposed methodology above was implemented in a case study based on the APS5000 all-electric APU structure and parameters. Model validation was conducted by comparing simulation results--covering steady-state, transients, healthy, and fault conditions--with reference data from third-party software and literature. The close agreement confirms both the model's accuracy and the effectiveness of our modeling methodology. This work establishes a modeling foundation for investigating the opportunities and challenges in fault detection and isolation (FDI) brought by the all electrification of the APU, including joint fault estimation and diagnosis, coupled electromechanical fault characteristics.

eess.SY↗

arXiv2025

Enhancing Fault Detection and Isolation in an All-Electric Auxiliary Power Unit (APU) Gas Generator by Utilizing Starter/Generator Signal

This study proposes a novel paradigm for enhancing fault detection and isolation (FDI) of gas generators in all-electric auxiliary power unit (APU) by utilizing shaft power information from the starter/generator. First, we conduct a pioneering investigation into the challenges and opportunities for FDI brought about by APU electrification. Our analysis reveals that the electrification of APU opens up new possibilities for utilizing shaft power estimates from starter/generator to improve gas generator FDI. We then provide comprehensive theoretical and analytical evidence demonstrating why, how, and to what extent, the shaft power information from the starter/generator can fundamentally enhance the estimation accuracy of system states and health parameters of the gas generator, while also identifying the key factors influencing these improvements in FDI performance. The effectiveness of the proposed paradigm and its theoretical foundations are validated through extensive Monte Carlo simulations. Furthermore, through comprehensive comparative analysis with state-of-the-art gas generator fault diagnosis methods, our experimental results not only demonstrate the superior performance of the proposed approach but also validate that the diagnostic capabilities of existing advanced FDI techniques can be substantially enhanced by incorporating shaft power information. And the observed performance improvement patterns strongly align with our theoretical analysis, verifying both the effectiveness and guiding significance of our theoretical framework. These research findings provide a unique perspective in answering three fundamental questions: why joint fault diagnosis of the starter/generator and gas generator is essential, how it can be implemented, and what factors determine its effectiveness, thereby opening up promising new avenues for FDI technologies in all-electric APU systems.

eess.SY↗

arXiv2025

Training-Free Guidance Beyond Differentiability: Scalable Path Steering with Tree Search in Diffusion and Flow Models

Training-free guidance enables controlled generation in diffusion and flow models, but most methods rely on gradients and assume differentiable objectives. This work focuses on training-free guidance addressing challenges from non-differentiable objectives and discrete data distributions. We propose TreeG: Tree Search-Based Path Steering Guidance, applicable to both continuous and discrete settings in diffusion and flow models. TreeG offers a unified framework for training-free guidance by proposing, evaluating, and selecting candidates at each step, enhanced with tree search over active paths and parallel exploration. We comprehensively investigate the design space of TreeG over the candidate proposal module and the evaluation function, instantiating TreeG into three novel algorithms. Our experiments show that TreeG consistently outperforms top guidance baselines in symbolic music generation, small molecule design, and enhancer DNA design with improvements of 29.01%, 16.6%, and 18.43%. Additionally, we identify an inference-time scaling law showing TreeG's scalability in inference-time computation.

cs.LG↗

arXiv2025

MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations

Large language models have demonstrated impressive performance on challenging mathematical reasoning tasks, which has triggered the discussion of whether the performance is achieved by true reasoning capability or memorization. To investigate this question, prior work has constructed mathematical benchmarks when questions undergo simple perturbations -- modifications that still preserve the underlying reasoning patterns of the solutions. However, no work has explored hard perturbations, which fundamentally change the nature of the problem so that the original solution steps do not apply. To bridge the gap, we construct MATH-P-Simple and MATH-P-Hard via simple perturbation and hard perturbation, respectively. Each consists of 279 perturbed math problems derived from level-5 (hardest) problems in the MATH dataset (Hendrycksmath et. al., 2021). We observe significant performance drops on MATH-P-Hard across various models, including o1-mini (-16.49%) and gemini-2.0-flash-thinking (-12.9%). We also raise concerns about a novel form of memorization where models blindly apply learned problem-solving skills without assessing their applicability to modified contexts. This issue is amplified when using original problems for in-context learning. We call for research efforts to address this challenge, which is critical for developing more robust and reliable reasoning models.

cs.LG↗

arXiv2024

Gradient Guidance for Diffusion Models: An Optimization Perspective

Diffusion models have demonstrated empirical successes in various applications and can be adapted to task-specific needs via guidance. This paper studies a form of gradient guidance for adapting a pre-trained diffusion model towards optimizing user-specified objectives. We establish a mathematical framework for guided diffusion to systematically study its optimization theory and algorithmic design. Our theoretical analysis spots a strong link between guided diffusion models and optimization: gradient-guided diffusion models are essentially sampling solutions to a regularized optimization problem, where the regularization is imposed by the pre-training data. As for guidance design, directly bringing in the gradient of an external objective function as guidance would jeopardize the structure in generated samples. We investigate a modified form of gradient guidance based on a forward prediction loss, which leverages the information in pre-trained score functions and provably preserves the latent structure. We further consider an iteratively fine-tuned version of gradient-guided diffusion where guidance and score network are both updated with newly generated samples. This process mimics a first-order optimization iteration in expectation, for which we proved O(1/K) convergence rate to the global optimum when the objective function is concave. Our code will be released at https://github.com/yukang123/GGDMOptim.git.

stat.ML↗