main concepts scientists sections
SimplePopularAdvanced

The Future File: Predicting Motion Without a Single Formula ✎ express

Original: "Machines that Predict Trajectories from Templates"
· Claudio De Persis, Pietro Tesi
arXiv:2607.11551 · 2026-07-13 · arXiv non-exclusive · independent analysis · 1 min
eess.SY cs.SY
Just gather an archive of past paths, and the computer, like a seasoned detective, will predict the continuation from the first steps.

📄 Showing the "Simple" version — "Popular" is not ready yet. Add it to favorites to help prioritize it.

An experienced detective doesn't calculate a criminal's path using the laws of physics: he checks a file of similar cases. The new prediction method works the same way: instead of complex equations, it has a vast archive of trajectories that have already happened — planetary motions, currency exchange rates, signals from space. Seeing the start of a new trajectory, the algorithm finds a similar snippet in the archive and fills in the continuation.

Detectors of gravitational waves catch the tremors of spacetime curvature exactly like this: the signal is compared against a library of half a million templates from mergers of black holes and neutron stars. This allowed Rainer Weiss and Barry Barish to hear the gravitational symphony.
And the most amazing part: if no template fits, then we're facing something unknown: this is how new physical phenomena can be discovered.

The method's authors went further: trajectory libraries can be combined, like crossbreeding plots, to produce behavior that never existed before. The system works with noise, errors, and without knowledge of laws. From controlling robots to climate prediction, the future file is ready to replace the oracle.

Whole graph →

The files of gravitational observatories already hold half a million templates, the largest collection of examples for predicting the future.

Scientists
Bernhard RiemannJoseph WeberKarl SchwarzschildKip ThorneRainer WeissAlbert Einstein
Concepts
predictive learning neural networks prediction Large Language Models training data protein language model neural networks reservoir computing predictive model Machine Learning classifier game theory molecular property prediction pattern recognition control theory probabilistic forecast template frame retrospective forecast Dynamical Systems dataset