Get Started
Home
Topics
Search
Library
Research questionHow can we predict micromagnetic magnetization trajectories efficiently over long horizons without accumulating rollout error?Repeated micromagnetic solver runs are expensive, while unconstrained reduced-order dynamics can drift as prediction rollouts extend. A useful predictor must preserve relevant precessional and dissipative behavior without requiring the full solver at every output time.
AI
Machine Learning
Neural and Evolutionary Computing
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.An Energy-Based Conservative-Dissipative Latent Neural Evolution Operator for Magnetization DynamicsThe source studies a convolutional autoencoder coupled to a structured latent neural ordinary differential equation, with an energy-based vector field containing antisymmetric and dissipative components. Models are trained on short trajectory windows using latent and decoded-rollout losses, without time-derivative supervision, physical-energy labels, or dissipation penalties. Evidence comes from two datasets parameterized by field amplitude and the two applied-field directions of NIST μMAG Standard Problem 4, with comparisons on short windows and uninterrupted rollouts extending to twice the training horizon against micromagnetic solver trajectories.research paper · Sep 3, 2026
Related questions
How can compact magnetic core-loss models adapt reconstructed hysteresis loops to operating conditions?How can longitudinal MRI models forecast neurodegenerative progression without confusing stable anatomy or noise with disease change?How can we simulate and backpropagate through long coupled dynamical systems without stepwise time recursion?How can on-policy diffusion RL reuse rollout computations without trading redundant recomputation for excessive memory use?