PROJECT STUDY / 03 / EXPERIMENTAL RESEARCH

AKASHA 2-Lite

Exploring what physics can bring to prediction.

Experiments comparing Hamiltonian dynamics models with unconstrained models on trajectories from physical systems.

The problem

A good short-term prediction can deteriorate when a model must reuse its own outputs. The project examines long-horizon prediction stability on simple physical systems.

The approach

The repository brings together experiment scripts, models, mathematical tests and reported results. Its central question compares Hamiltonian and unconstrained dynamics, including pendulum and harmonic oscillator experiments.

Status & limitations

Exploratory work; no independent validation is presented here. Conservation constraints are not appropriate for every system: the repository’s table reports a better baseline result for the damped pendulum. Interactive demos are separate from experimental validation.

Let’s explore

Baseline selection, training / evaluation separation, loss-function effects and the limits of physical inductive biases.

LET’S BUILD WHAT COMES NEXT

A complex idea.
A simple conversation.

Let’s talk about agents, products and what we could build together.

Get in touch