Selected Projects in Machine Learning Engineering and Data Science

---

Transformer for Global, Multi-horizon Time Series Forecasting on TPUs

I implemented a Vanilla Transformer for Time Series Forecasting on Google Cloud TPU accelerators. I trained this architecture on two standard datasets for global, multi-horizon forecasting (electricity and traffic) and achieved training time reductions from several hours to under two minutes. This research achievement was published as García-Nava et al., 2022 by Springer-Nature’s The Journal of Supercomputing.

View code on GitHub

View research paper on The Journal of Supercomputing

---

Deep Learning Architectures for Electric Power Forecasting on Google Cloud Tensor Processing Units

I implemented a deep multi-sequence stacked LSTM and an encoder-decoder with attention architectures for electric load short-term forecasting on Google Cloud TPU accelerators. The models achieved outstanding predictive performance and training time reductions from several hours to under 30 seconds wall-time.

View code on GitHub

---

Mobile Application for Walking Routes Tracking (Android)

I implemented an Android application for tracking and cloud storing walking routes in public spaces of Morelia city. Walking routes and points of interest are persisted in Firebase for further walk-ability analysis.

View code on GitHub

---

ML Pipeline for Large Scale Power Quality Forecasting with Apache Spark and SciKit-Learn

I designed a machine learning pipeline for large scale power quality short-term forecasting in Central-West CFE (state-owned Mexican grid operator) distribution network. This pipeline executes more than 4,000 forecasting models on a weekly basis since 2019.

View research paper on Applied Sciences

---

Page template forked from evanca