JNLR
JAX-based non-linear reconciliation and learning
J-NLR is a Python library for non-linear reconciliation, learning, and geometric analysis on constraint manifolds. Built on JAX, it leverages automatic differentiation and GPU/TPU acceleration to efficiently project predicted values onto surfaces defined by implicit constraints \(f(z) = 0\).
Example notebooks — geodesics, meshes, projection, and sampling (scripts/render_notebook_spin_reel.py)
Key Features
- Non-linear Reconciliation: Multiple solvers (Augmented Lagrangian, curvature-aware Newton, vanilla projections) for projecting forecasts onto constraint manifolds
- SHOULD Analysis: Curvature-based methods to determine when reconciliation is beneficial—verify if RMSE is guaranteed to reduce before applying corrections
- Manifold Sampling: Sample from explicit (graph) or implicit manifolds using volume-weighted sampling, Latin hypercube, or Langevin dynamics on the constraint surface
- Mesh Generation: Create triangulated meshes from explicit parameterizations for visualization and geodesic computation
- Geodesics: Compute geodesic distances and shortest paths on manifolds via exact MMP algorithm or fast graph-based approximations; includes probabilistic scores like pointcloud geodesic distance
- Visualization: Interactive 3D rendering of manifolds, projections, and geodesic paths with Plotly
- JAX-native: Fully JIT-compiled and vectorized (
vmap) for high-performance batch processing
API Documentation
Explore the full API reference:
- Reconcile - Non-linear reconciliation solvers
- Should - SHOULD analysis for curvature-based decision making
- Stats - Statistical utilities
- Curvature Utils - Curvature computation utilities
Installation
Install using uv package manager:
uv pip install -e .
Or with pip:
pip install jnlr
Citation
If you use JNLR in academic work, please cite the associated paper:
Lorenzo Nespoli, Anubhab Biswas, Roberto Rocchetta, and Vasco Medici.
"Nonlinear reconciliation: Error reduction theorems."
Transactions on Machine Learning Research (TMLR), 2026.
OpenReview: https://openreview.net/forum?id=dXRWuogm3J
BibTeX
@article{nespoli2026nonlinear_reconciliation,
title = {Nonlinear reconciliation: Error reduction theorems},
author = {Nespoli, Lorenzo and Biswas, Anubhab and Rocchetta, Roberto and Medici, Vasco},
journal = {Transactions on Machine Learning Research},
year = {2026},
url = {https://openreview.net/forum?id=dXRWuogm3J},
note = {Accepted by TMLR}
}
Acknowledgements
This work has been funded by the Swiss State Secretariat for Education, Research and Innovation (SERI) under the Swiss contribution to the Horizon Europe projects DR-RISE (Horizon Europe, Grant Agreement No. 101104154) and REEFLEX (Horizon Europe, Grant Agreement No. 101096192).