Generative Adaptation of Dynamics to Environmental Shifts via Weight-space Diffusion

2025/05/20 by Ruikun Li, Li, Ruikun, Huandong Wang +9 · 2 citations
Computer Science · Physics and Astronomy · #Computational Engineering #FOS: Computer and information sciences #Finance #Gaussian Processes and Bayesian Inference #Generative Adversarial Networks and Image Synthesis #Model Reduction and Neural Networks #and Science (cs.CE)

paper · pdf · doi:10.48550/arxiv.2505.13919

openalex publication_date 2025/05/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Data-driven dynamics prediction often fails under environmental shifts, while traditional fine-tuning remains computationally prohibitive for hardware-constrained or data-scarce applications. We propose DynaDiff, a generative meta-learning framework that transitions the paradigm from gradient-based tuning or modulation to direct weight-space generation. Specifically, we first abstract expert weights as novel weight graphs, utilizing multi-head attention to explicitly capture topological coupling within weights. Subsequently, we design a functional loss to ensure that the generated models achieve consistency with expert models in physical behavior. Finally, we develop a dynamics-informed prompter that extracts cross-domain physical and spectral features from observation sequences to condition the diffusion model. Experiments demonstrate that DynaDiff boosts average prediction accuracy by 10.78% over competitive baselines. Furthermore, by pre-constructing a model zoo of expert predictors, we amortize the fine-tuning overhead into a one-time offline cost, significantly boosting deployment efficiency in new environments.

Citations

• Zero-Shot Forecasting of Network Dynamics through Weight Flow Matching
• Text2Weight: Bridging Natural Language and Neural Network Weight Spaces
• WeightFlow: Learning Stochastic Dynamics via Evolving Weight of Neural Network
• Continual Adaptation: Environment-Conditional Parameter Generation for Object Detection in Dynamic Scenarios
• Drag-and-Drop LLMs: Zero-Shot Prompt-to-Weights
• LoRA-Gen: Specializing Large Language Model via Online LoRA Generation
• Text-to-LoRA: Instant Transformer Adaption
• Beyond Equilibrium: Non-Equilibrium Foundations Should Underpin Generative Processes in Complex Dynamical Systems
• Sparse Diffusion Autoencoder for Test-time Adapting Prediction of Complex Systems
• Predicting the Dynamics of Complex System via Multiscale Diffusion Autoencoder
• Shape Generation via Weight Space Learning
• Structure Is Not Enough: Leveraging Behavior for Neural Network Weight Reconstruction
• In-Context Meta LoRA Generation
• Recurrent Diffusion for Large-Scale Parameter Generation
• Efficient Training with Denoised Neural Weights
• Model Zoo Dataset Samples for Scalable Weight Space Learning
• Poseidon: Efficient Foundation Models for PDEs
• Neural Context Flows for Meta-Learning of Dynamical Systems
• Pretraining Codomain Attention Neural Operators for Solving Multiphysics PDEs
• Graph Neural Networks for Learning Equivariant Representations of Neural Networks
• DPOT: Auto-Regressive Denoising Operator Transformer for Large-Scale PDE Pre-Training
• Diffusion-Based Neural Network Weights Generation
• Data-Efficient Operator Learning via Unsupervised Pretraining and In-Context Learning
• Neural Network Diffusion
• Spatio-Temporal Few-Shot Learning via Diffusive Neural Network Generation
• Neural Operators for Accelerating Scientific Simulations and Design
• Scientific discovery in the age of artificial intelligence
• MetaDiff: Meta-Learning with Conditional Diffusion for Few-Shot Learning
• Accurate medium-range global weather forecasting with 3D neural networks
• HyperDiffusion: Generating Implicit Neural Fields with Weight-Space Diffusion
• Scalable Diffusion Models with Transformers
• Plug-and-Play Diffusion Features for Text-Driven Image-to-Image Translation
• PDEBENCH: An Extensive Benchmark for Scientific Machine Learning
• Hyper-Representations as Generative Models: Sampling Unseen Neural Network Weights
• Diffusion Models in Vision: A Survey
• U-NO: U-shaped Neural Operators
• Generalizing to New Physical Systems via Context-Informed Dynamics Model
• From data to functa: Your data point is a function and you can treat it like one
• High-Resolution Image Synthesis with Latent Diffusion Models
• High-Resolution Image Synthesis with Latent Diffusion Models
• Diffusion Autoencoders: Toward a Meaningful and Decodable Representation
• LEADS: Learning Dynamical Systems that Generalize Across Environments
• Diffusion Models Beat GANs on Image Synthesis
• Meta-Learning Dynamics Forecasting Using Task Inference
• Fourier Neural Operator for Parametric Partial Differential Equations
• Multiscale Simulations of Complex Systems by Learning their Effective Dynamics
• Denoising Diffusion Probabilistic Models
• Modal Analysis of Fluid Flows: Applications and Outlook
• Learning Dynamical Systems from Partial Observations
• Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
• Theoretical Analysis of Domain Adaptation with Optimal Transport
• OmniArch: Building Foundation Model For Scientific Computing

Cited by

• Zero-Shot Forecasting of Network Dynamics through Weight Flow Matching
• WeightFlow: Learning Stochastic Dynamics via Evolving Weight of Neural Network
• Predicting the Dynamics of Complex System via Multiscale Diffusion Autoencoder
• MLLM-based Discovery of Intrinsic Coordinates and Governing Equations from High-Dimensional Data

Related

• Environment Inference for Learning Generalizable Dynamical System
• Consistent World Models via Foresight Diffusion
• LEADS: Learning Dynamical Systems that Generalize Across Environments
• Predicting partially observable dynamical systems via diffusion models with a multiscale inference scheme
• Dynamical Diffusion: Learning Temporal Dynamics with Diffusion Models