Rae Chipera

About

Hi. I'm a doctoral candidate at National University working on reservoir computing and nonlinear dynamics. My dissertation investigates fractal activation functions in Echo State Networks: what happens when you replace the usual smooth nonlinearities with something mathematically wilder.

In other words, I build machine learning models that don't break when the data is chaotic or messy.

Before this, I worked in Marine Corps Intelligence, deployed to Iraq, trained in Arabic at DLI, and spent a while in quantitative finance. I taught myself calculus partly while deployed and partly in a national lab library. The path here was not a straight line.

I also run Jaxorik AI Research Group, a New Mexico firm focused on explainable AI for disaster response and critical infrastructure.

Reservoir computing Fractal geometry Echo State Networks Explainable AI Critical infrastructure Nonlinear dynamics

Currently

Spring / Summer 2026. Proposal defended in April. Filing IRB exemption, finishing empirical runs on CIFAR-10, Galaxy Zoo 2, and Gravity Spy (LIGO detector glitches). Prepping a manuscript for submission to Neural Networks. Always happy to hear from people working on similar problems — see Contact.

Research

Beyond Lipschitz Continuity and Monotonicity: Fractal and Chaotic Activation Functions in Echo State Networks preprint

with J. Du & I. Tsapara · arXiv:2512.14675

Theoretical analysis of fractal activation functions in Echo State Networks, introducing the Degenerate Echo State Property (d-ESP) as a formal framework for characterizing reservoir dynamics under non-smooth, self-similar activations. Analytical work; empirical validation is the dissertation.

Beyond Smooth Activations: Irregular Functions for Modeling Chaotic Data Patterns in Neural Networks dissertation · in progress

National University · Committee: Du, Tsapara, Dhou · proposal defended April 2026

Empirical study of Cantor, Weierstrass, Logistic Map, and Mandelbrot activations across three benchmarks: CIFAR-10 (control), Galaxy Zoo 2 (morphological classification), and Gravity Spy (LIGO interferometer glitch detection). MANOVA-based statistical design. The goal is a principled theoretical account of when and why fractals work in recurrent architectures.

Code

fractal_reservoir releases with journal submission

Python · dissertation codebase

ESN implementation with fractal activation functions. Includes Cantor, Weierstrass, and Mandelbrot variants, memory capacity benchmarks (following Jaeger 2001), and experiment pipelines for CIFAR-10, GZ2, and Gravity Spy.

PyRae

Python · from-scratch deep learning

NumPy-based PyTorch re-implementation. Good for understanding what's actually happening under the black box.

satellite-anomaly-detection-demo

Python · Jaxorik demo

Anomaly detection pipeline for satellite data, built as a technical demonstration for Jaxorik's explainable AI work in disaster response and critical infrastructure.

Teaching

Private Mathematics Instruction ongoing

Rio Rancho, NM

One-on-one instruction in Geometry, Algebra II, AP Calculus, AP Statistics, and college-level Statistics.

Contact

Best reached by email at rachipe+web@jaxorik.com. I reply to real emails from real people; please write one.

Especially interested in hearing from:

Researchers working on nonlinear dynamics in reservoir computing. Theory, architectures, training methods, anything adjacent. If you've thought seriously about why ESNs do what they do, I want to talk.

People with "fussy" datasets that resist clean modeling. Plasma physics, seismology and plate tectonics, space weather, turbulent flow, neural recordings — anything chaotic, non-stationary, or otherwise allergic to standard architectures. I'd like to test fractal-activation ESNs on your data and see what happens.

What you get if we collaborate: I do the modeling work, help write up the results, and stay non-flaky through to publication. You get full access to the code and trained models. Transparency is the whole point of research collaboration, so it would be weird to hoard the artifacts.

If you're with a government agency or federal contractor: please route inquiries through my company instead. Jaxorik AI Research Group is registered on SAM.gov (UEI K8ENCCGZ2M13) as WOSB/SDVOSB and is set up to handle the contracting side properly.