Research¶
nefi is the software form of a series of papers on neural-field inversion from the CAB Lab at Princeton University. Two are published so far, NeTMY and NeFTY. The library keeps their methods as defaults, their numbers as configuration fields, their explanations as diagnostics and their evaluation protocol as a benchmark harness, and generalizes the recipe to any differentiable instrument. Later papers in the series are added the same way.
-
NeTMY (NV-center noise sensing) and NeFTY (thermal tomography): what each one solves, what it contributes, what nefi implements, and how to cite them.
-
The representation is the prior
A parameterized update is the raw gradient filtered by \(G_\theta = J_\theta J_\theta^\top\). Why that single fact explains center collapse, spectral bias and the choice of representation — and how nefi measures it.
-
The model-error floor of the heat operator, when a TV grid matches the neural field, what an accepted refinement certifies, and what tuning cannot supply.
-
Every equation, proposition, table and key figure of both papers, mapped to the code that implements, measures or reproduces it.
-
The architecture contract: design principles, package layout and the exact signatures every module builds against.
-
The server runbook: exact commands for the headline tables, their cost in GPU-hours, and the caveats to check first.