Alejandro Cano.

eX Virentia (eXVi) explores how existing, unmodified vegetation can act as a sensor for activities of Department of Defense interest by using natural plant responses to man-made chemical exposures.

The program brings plant biologists together with remote-sensing and modeling experts to study biochemical responses to chemical exposure and their externally observable characteristics. I provide computational and graph-theoretic support to Purdue’s College of Agriculture Research Services team.

This project addresses the need of the aerospace propulsion system Prognostics & Health Management (PHM) team for synthetic data that mimics actual fleet data. The data can be shared with external research entities, including universities, to explore prognostic algorithms while respecting export and International Traffic in Arms Regulations (ITAR) restrictions on actual data.

I spent one semester organizing a student team and co-developing machine-learning models to support the Rolls-Royce PHM team.

A basic two-layer neural network classifies handwritten MNIST digits without PyTorch or TensorFlow. It uses NumPy and Pandas, ReLU activations, softmax, backpropagation, and gradient descent. The original page reports accuracy above 91% after fewer than 500 iterations.

The implementation loads the MNIST CSV, normalizes grayscale values, trains a 784→10→10 network, evaluates on a development split, and plots sample predictions.

A PyTorch Wasserstein Generative Adversarial Network generates handwritten images from MNIST. The implementation includes a generator, critic, gradient penalty, training loop, saved generated samples, and training-loss curves.

This project solves a dynamic optimal-transport problem with a Neural ODE. A Gaussian distribution is transported toward a distribution sampled from an MNIST digit using a continuous normalizing flow, kinetic-energy regularization, and trajectory visualizations.