https://surrogatemodels.com/ News, papers, datasets, and funding and deals in Physics AI for engineers: General physics. en https://surrogatemodels.com/papers/neural-odehttps://surrogatemodels.com/papers/neural-odeThu, 12 Nov 2026 09:00:00 GMTA neural network gives the time derivative of a hidden state, and a black-box ODE solver computes the output. Gradients come from the adjoint sensitivity method, a second ODE solved backwards in time, so training does not store the solver's intermediate steps.General physics https://surrogatemodels.com/papers/sindyhttps://surrogatemodels.com/papers/sindyThu, 12 Nov 2026 09:00:00 GMTStacks time histories of a system's state and of its time derivative, builds a library of candidate nonlinear functions of the state (polynomials, for example) and uses sparse regression to keep the fewest library terms that reproduce the derivative. The only assumption is that the governing equations have few terms.General physics https://surrogatemodels.com/papers/ai-feynmanhttps://surrogatemodels.com/papers/ai-feynmanThu, 12 Nov 2026 09:00:00 GMTThe algorithm looks for a formula that fits a table of data. It first fits a neural network to the data, uses the network to test for symmetries and separability, and so splits the problem into smaller ones with fewer variables. It combines this with dimensional analysis, polynomial fits and brute-force search.General physics https://surrogatemodels.com/papers/hnnhttps://surrogatemodels.com/papers/hnnThu, 12 Nov 2026 09:00:00 GMTThe network outputs one scalar, a learned Hamiltonian, from positions and momenta. The time derivatives are taken from its gradient through Hamilton's equations, so the model conserves a quantity analogous to total energy by construction.General physics