22 papers
SR-OPSD: Self-Referenced On-Policy Self-Distillation
Zhuo Sun, Entong Li, Yanlong Zhao +9
On-policy self-distillation (OPSD) converts feedback into dense token-level supervision on student-generated trajectories, complementing reinforcement learning with sparse outcome…
From Deterministic to Generative Deep Learning for Urban Air Quality Reconstruction from Sparse Observations
Abhishek A. Sabnis, Mihai Mitrea, Lya Lugon +5
Full-field reconstruction of air pollution is essential for evaluating pollution exposure and supporting public health decision-making. However, the complex interactions among poll…
NEUROSYMLAND: Neuro-Symbolic Landing-Site Assessment for Robust and Edge-Deployable UAV Autonomy
Weixian Qian, Tianyi Yang, Sebastian Schroder +5
Safe landing-site assessment in unstructured environments remains a key challenge for autonomous UAV deployment, as vision-only learning approaches often degrade under terrain vari…
A Gradient Flow Perspective on Minimum MMD Estimation
Sophia Seulkee Kang, Louis Sharrock, Xiaoyuan Cheng +2
Minimum maximum mean discrepancy (MMD) estimation has emerged as a robust and likelihood-free alternative to maximum likelihood estimation for parameter estimation. Yet, despite it…
Scaling World-Model Reinforcement Learning Through Diffusion Policy Optimization
Xiaoyuan Cheng, Wenxuan Yuan, Zhancun Mu +5
Model-based reinforcement learning (RL) can be effectively supported at scale through the use of world models. However, in practice, scaling such approaches remains fundamentally l…
Saliency-Aware Regularized Quantization Calibration for Large Language Models
Yanlong Zhao, Xiaoyuan Cheng, Huihang Liu +6
Post-training quantization (PTQ) is an effective approach for deploying large language models (LLMs) under memory and latency constraints. Most existing PTQ methods determine quant…
Fisher Decorator: Refining Flow Policy via a Local Transport Map
Xiaoyuan Cheng, Haoyu Wang, Wenxuan Yuan +4
Recent advances in flow-based offline reinforcement learning (RL) have achieved strong performance by parameterizing policies via flow matching. However, they still face critical t…
From Particles to Perils: SVGD-Based Hazardous Scenario Generation for Autonomous Driving Systems Testing
Linfeng Liang, Xiao Cheng, Tsong Yueh Chen +1
Simulation-based testing of autonomous driving systems (ADS) must uncover realistic and diverse failures in dense, heterogeneous traffic. However, existing search-based seeding met…
Spatiotemporal System Forecasting with Irregular Time Steps via Masked Autoencoder
Kewei Zhu, Yanze Xin, Jinwei Hu +3
Predicting high-dimensional dynamical systems with irregular time steps presents significant challenges for current data-driven algorithms. These irregularities arise from missing…
How Does the Lagrangian Guide Safe Reinforcement Learning through Diffusion Models?
Xiaoyuan Cheng, Wenxuan Yuan, Boyang Li +7
Diffusion policy sampling enables reinforcement learning (RL) to represent multimodal action distributions beyond suboptimal unimodal Gaussian policies. However, existing diffusion…
Outlier-robust Diffusion Posterior Sampling for Bayesian Inverse Problems
Yiming Yang, Xiaoyuan Cheng, Yi He +3
Diffusion models have emerged as powerful learned priors for Bayesian inverse problems (BIPs). Diffusion-based solvers rely on a presumed likelihood for the observations in BIPs to…
Planning with Language and Generative Models: Toward General Reward-Guided Wireless Network Design
Chenyang Yuan, Xiaoyuan Cheng
Intelligent access point (AP) deployment remains challenging in next-generation wireless networks due to complex indoor geometries and signal propagation. We firstly benchmark gene…
Generative Modeling through Koopman Spectral Analysis: An Operator-Theoretic Perspective
Yuanchao Xu, Fengyi Li, Masahiro Fujisawa +3
We propose Koopman Spectral Wasserstein Gradient Descent (KSWGD), a particle-based generative modeling framework that learns the Langevin generator via Koopman theory and integrate…
From Embedding to Control: Representations for Stochastic Multi-Object Systems
Xiaoyuan Cheng, Yiming Yang, Wei Jiang +3
This paper studies how to achieve accurate modeling and effective control in stochastic nonlinear dynamics with multiple interacting objects. However, non-uniform interactions and…
Information Shapes Koopman Representation
Xiaoyuan Cheng, Wenxuan Yuan, Yiming Yang +4
The Koopman operator provides a powerful framework for modeling dynamical systems and has attracted growing interest from the machine learning community. However, its infinite-dime…
Safe and Stable Control via Lyapunov-Guided Diffusion Models
Xiaoyuan Cheng, Xiaohang Tang, Yiming Yang
Diffusion models have made significant strides in recent years, exhibiting strong generalization capabilities in planning and control tasks. However, most diffusion-based policies…
Fast-Forward Lattice Boltzmann: Learning Kinetic Behaviour with Physics-Informed Neural Operators
Xiao Xue, Marco F. P. ten Eikelder, Mingyang Gao +7
The lattice Boltzmann equation (LBE), rooted in kinetic theory, provides a powerful framework for capturing complex flow behaviour by describing the evolution of single-particle di…
Chaos Meets Attention: Transformers for Large-Scale Dynamical Prediction
Yi He, Yiming Yang, Xiaoyuan Cheng +4
Generating long-term trajectories of dissipative chaotic systems autoregressively is a highly challenging task. The inherent positive Lyapunov exponents amplify prediction errors o…
Learning Chaos In A Linear Way
Xiaoyuan Cheng, Yi He, Yiming Yang +5
Learning long-term behaviors in chaotic dynamical systems, such as turbulent flows and climate modelling, is challenging due to their inherent instability and unpredictability. The…
Tensor-Var: Efficient Four-Dimensional Variational Data Assimilation
Yiming Yang, Xiaoyuan Cheng, Daniel Giles +5
Variational data assimilation estimates the dynamical system states by minimizing a cost function that fits the numerical models with the observational data. Although four-dimensio…
KEEC: Koopman Embedded Equivariant Control
Xiaoyuan Cheng, Yiming Yang, Xiaohang Tang +2
An efficient way to control systems with unknown nonlinear dynamics is to find an appropriate embedding or representation for simplified approximation (e.g. linearization), which f…
Safe Reinforcement Learning in Tensor Reproducing Kernel Hilbert Space
Xiaoyuan Cheng, Boli Chen, Liz Varga +1
This paper delves into the problem of safe reinforcement learning (RL) in a partially observable environment with the aim of achieving safe-reachability objectives. In traditional…