Skip to main content
Event

AI Latin American SumMIT Hackathon 2020

Dr. Omar Costilla Reyes organized the AI Latin American SumMIT Hackathon at the MIT Media Lab on January 24–25, 2020, where he also taught workshops. The workshops covered deep learning, PyTorch and gradient-based learning.

Poster announcing the AI Latin American SumMIT Hackathon at MIT, January 24–25, 2020. Credit: Image courtesy of Dr. Omar Costilla Reyes. View image source.
Workshop 1, an introduction to deep learning and PyTorch, at the AI Latin American SumMIT Hackathon, January 24, 2020. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Workshop 2, on optimization and gradient-based learning, at the AI Latin American SumMIT Hackathon, January 24, 2020. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Teams preparing their final projects on the hackathon’s second day, January 25, 2020. Credit: Image courtesy of Dr. Omar Costilla Reyes.
Teams work in the open space of the MIT Media Lab during the hackathon, January 24, 2020. Credit: Image courtesy of Dr. Omar Costilla Reyes.

A hackathon after the summit

Dr. Omar Costilla Reyes organized the AI Latin American SumMIT Hackathon, a two-day event held at the MIT Media Lab on January 24 and 25, 2020. It followed the AI Latin American SumMIT, held at the Media Lab on January 21–23, which MIT News described as an effort by four MIT graduate students, Dr. Costilla Reyes among them, to bring Latin American policymakers and AI researchers together to speed up the adoption of AI in the region. A hackathon is a short, intensive event in which teams build a working prototype around a shared problem. The hackathon poster names five themes drawn from the United Nations Sustainable Development Goals: no poverty; quality education; peace, justice and strong institutions; climate change and sustainable development; and good health and wellbeing.

Workshops on how models learn

Speakers Pablo Samuel Castro and Andreas Müller led a machine-learning workshop during the hackathon, and Dr. Costilla Reyes taught workshops on deep learning, neural networks, PyTorch, optimization, and gradient-based learning. A neural network is a model built from layers of simple mathematical functions whose parameters are adjusted from examples; PyTorch is an open-source software library for building and training such models. Gradient-based learning is the usual way the adjustment is done: the model’s error on the training data is measured, and each parameter is nudged in the direction that reduces it, over many repetitions. On January 24 the workshops began with an introduction to deep learning and PyTorch, and a second session on optimization and gradient-based learning followed. In the first, participants also discussed the difference between linear and nonlinear models, that is, between models that can only express straight-line relationships in the data and models that can capture curved, more complex ones.

Teams and final projects

The two days combined this teaching with project work. On the second and final day, January 25, teams were getting ready to present their final projects, and the photographs on this page show the workshop rooms and the open workspace where they worked. The summit itself, including the Day 2 panel on AI and the Sustainable Development Goals that Dr. Costilla Reyes moderated, is described on the AI Latin American SumMIT page.

Topics