Get Started
Home
Topics
Search
Library
Research questionHow can we simulate and backpropagate through long coupled dynamical systems without stepwise time recursion?Time-marching simulation computes each state from the previous one, so long coupled systems create substantial sequential depth. Backpropagation through those steps also inherits depth-dependent memory and recursion.
AI
Machine Learning
Neural and Evolutionary Computing
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.Time Without Timesteps: Simulating Coupled Dynamical Systems via Self-ConsistencyThe evidence concerns learned surrogate operators for coupled van der Pol oscillators and Hodgkin–Huxley neuron networks. Each surrogate maps a full driving trajectory and initial condition to a full output trajectory, while coupling is enforced as a fixed-point problem over complete trajectories. Reported results compare 4–10 Newton iterations with 1,500-step reference integration and examine convergence through the surrogate Jacobian’s spectral radius; they also show that surrogate error can degrade the approach.research paper · Sep 3, 2026
Related questions
How can unknown nonlinear ODE dynamics and time-varying external forces be learned from one trajectory without physics priors?How can learned graph simulators preserve contact history for accurate long-horizon granular-dynamics rollouts?How can periodic orbits in dynamical systems be continued and bifurcations detected without time integration or hand-derived Jacobians?How can GPU ODE ensemble solvers provide fast reverse-mode gradients without sacrificing solve speed?