SciML / SciML/NeuralOperators.jl
Implement Spectral Neural Operator (SNO)
Nobody has claimed this yet.
- Dominant language
- Julia
- Stars
- 41
- Forks
- 15
- Avg merge
- 13h 14m
- Merged PRs (30d)
- 12
Description
Summary
Implement SNO, which maps between spectral coefficients (Chebyshev, Legendre, or Fourier) and supports multiple polynomial basis choices.
Reference
- Fanaskov & Oseledets, "Spectral Neural Operators," Doklady Mathematics, 2023. arXiv:2205.10573
Description
SNO operates directly on spectral expansion coefficients, mapping between Chebyshev, Legendre, or Fourier series coefficients of the input and output functions. Unlike FNO which truncates modes, SNO works with the full spectral representation and avoids aliasing. The paper reports superior performance to FNO and DeepONet for many operators.
This would require:
- New transform types (Chebyshev, Legendre) implementing the
AbstractTransforminterface - Coefficient-space linear layers
- Integration with existing architecture patterns
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reading the existing architecture patterns and the AbstractTransform interface. Use Fanaskov & Oseledets, “Spectral Neural Operators,” as the behavioral reference, then determine how Chebyshev, Legendre, and Fourier coefficient mappings fit the project. Done means the new transforms and coefficient-space layers are integrated into the existing architecture.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- julia
- Domain
- machine-learning
- Issue type
- Feature
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 28/100