Skip to content

Concepts

The library's design, from the abstraction every model shares to the conventions that make them interchangeable.

  • INRs in Inro

    A model is a function of coordinates, a dataset is a set of coordinate/value pairs, and inro.train ties them together.

  • FUTON

    The Fourier Tensor Network, stage by stage: bases, combiners, heads, and how to size a model.

  • Error bound of FUTON

    How close a FUTON with a linear head can come to a signal, exactly and in computable bounds, with the proof and an experiment.

  • The model zoo

    The twelve architectures of the benchmarks, their families, references and constructor arguments.

  • Conventions

    Notation, coordinate and value ranges, tensor shapes, metrics, checkpoints and seeds.