Install via pip install jnlr or uv add jnlr
GitHub: github.com/supsi-dacd-i...
Documentation: supsi-dacd-isaac.github.io/JNLR/

The main capabilities of the library:
* projection on arbitrary explicit (charts) or implicit manifolds 📐
* mesh generation on 3D manifolds 🌐
* geodesic paths and geodesic distance computation 🐜
* sampling of explicit/implicit manifolds 🥏
* interactive visualisation 🧿

Capabilities of JNLR: geodesics, mesh rendering, projection, sampling

J-NLR is a JAX native Python library for non-linear reconciliation, learning, and geometric analysis on constraint manifolds. Check this out!

How AI slop is causing a crisis in computer science Preprint repositories and conference organizers are having to counter a tide of ‘AI slop’ submissions.

How do we fix this problem?
Simply we don't have enough reviewers, and AI slop can sound very convincing.
Push for more code to be open sourced like AAAI? More stringent reproducibility requirements like the international journal of forecasting?

www.nature.com/articles/d41...

When compared to sizing using a prescient MPC (controller with perfect forecasts), the method provides smaller system sizing, since it considers more realistic operation performances, but with similar LCOE. For more details and plots: arxiv.org/pdf/2511.21619

Furthermore we tune it to minimize CVaR of (past!) daily peaks on a training set. This provides:
* Interpretable controller
* No need for forecasts
* Conservative control
* Super fast, so that we can optimize for LCOE of the system considering control

Turns out that RL is not much different from parametric rule based controllers (RBC), as they're both stochastic optimizations at their core. Following ideas from Powell, we propose a simple tunable RBC based on running quantiles of the demand.

Shaving peaks of demand operating a battery is a relatively difficult task, which usually involves forecasting future peaks.

Model predictive control, reinforcement learning (RL) and multistage stochastic control has been used among others to optimize this particular problem.

I'm presenting work at the 2026 IEEE Power & Energy Society General Meeting, join me Learn more

Wow! IEEE PES General Meeting has a greeting card to shill your accepted work!

Well.. Let me take the opportunity to share our paper:

Robust Rule-Based Sizing and Control of Batteries
for Peak Shaving Applications

invt.io/1bxbbpco1gp

Agree, no substantial leakage is involved and it's less prone to fluctuation in the data

For the LeJEPA/SIGReg/SSL demo, I added a summary of the algorithm (scroll down) and links to more resources: www.scotthawley.com/ssltoy