AMLCville Retrospective

On April 12th, 2018, Charlottesville and the Tom Tom Founders Festival played host to the second annual Applied Machine Learning Conference (AMLCville). I had the fortune to attend AMLCville and listen to talks on a wide range of topics and get to see the amazing ways that machine learning is being used out in the world. AMLCville was split into three different tracks with topic sessions on natural language processing, health care, computer vision, and geospatial analysis.

I spent most of my time in the general interest sessions and attended the keynotes. The two big highlights for me were the keynote given by David Luebke of Nvidia on Deep Learning and Computer Graphics and the talk by Renée Teate of HelioCampus on bias in machine learning titled Can a Machine be Racist or Sexist?

Title slide from David Luebke’s keynote Deep Learning and Computer Graphics

Title slide from David Luebke’s keynote Deep Learning and Computer Graphics

David Luebke’s keynote gave an overview of the work Nvidia has been doing using deep learning for computer graphics. One of the examples that stood out to me was producing new mouth animations using audio files. By training a deep learning model using audio and motion capture, they were able to produce new mouth animations using only new audio files. While the new mouth animations weren’t perfect, they went a long way to making the animated figure look like they were saying the words coming from the audio. I can’t wait to see how this technology could be applied to games in the future. I can imagine that localizing games into new languages would benefit from the technology and hopefully lower the bar for localizations.

Title slide from Renée Teate’s talk Can a Machine be Racist or Sexist?

Title slide from Renée Teate’s talk Can a Machine be Racist or Sexist?

Renée Teate’s talk gave an overview of the different ways that bias can impact machine learning. After her talk, Renée Teate moderated a panel discussion on bias in machine learning with a few of the other speakers that day. I think my favorite line came during the panel in which one of the panelists (Ines Montani I believe) reframed the often quoted phrase “your model is only as good as your data” as “your model can often be much worse than your data.” A good reminder that, when building models, we need to make sure the model makes sense in addition to the data being good.

I’m quite excited to see data science and machine learning grow in Charlottesville. Since graduating and starting to dive into the data science community, it has been incredibly useful to have such a great resource within arms reach. Thank you to all of the organizers and speakers for putting on such a great conference!