Skip to main content

One differentiable model, many tasks

Simulation, estimation, reconciliation, design of experiments, optimization, control and planning are different questions asked of the same differentiable model f(x; θ).

Open the graphic full-size.

Why differentiable models?

If there is one nearly unifying principle in optimization, it is that derivatives play a central role in much of it (I know not all of it; integers are not differentiable, and you often have to estimate derivatives, etc.). Optimization means many things as illustrated above. We can use it to fit parameters, simulate, reconcile data, plan, control, design experiments and more! These all rely on the central model, and different combinations of what know, what you want, and how you get it. What we aim for in our ecosystem shown above is for a single model to be reusable across many tasks.

The graphic above is interactive, and you can see how the task is implemented in many of the tasks through notebook examples. We have not fully integrated everything seamlessly yet, but we are working on it and making good progress. If you find bugs or have suggestions, please let us know by opening an issue on GitHub!