Making deep learning accessible

The Data Exchange Podcast: Piero Molino on opening up deep learning to non experts.

Subscribe: AppleAndroidSpotifyStitcherGoogleRSS.

In this episode of the Data Exchange I speak with Piero Molino, creator of Ludwig, a toolbox that allows users to train and test deep learning models through a declarative interface. Piero created Ludwig while serving as a Senior Research Scientist at Uber AI. He originally created Ludwig for his personal use and it slowly garnered users within Uber. By the time it was open sourced in early 2019, the project immediately found a receptive audience in the conferences I was chairing at the time.

Download the 2020 NLP Survey Report and learn how companies are using and implementing natural language technologies.

Today the project has 7,000 stars on Github and it continues to “make deep learning easier to understand for non-experts and enable faster model improvement iteration cycles for experienced machine learning developers and researchers alike”.

Our conversation touched on many aspects of the project including the recent 0.3 release. With their upcoming 0.4 release, Piero shared their plans to have more integrations with Ray:

    There are at least three things that we want to do with Ray. … First is distributed training using Horovod’s integration with Ray. … Second we want to integrate Ray Tune into our hyperparameter optimization tools. … Finally we also want to be able run data preprocessing in a parallel way on a Ray cluster.

Actually Ludwig is just one of a growing number of popular data and machine learning libraries that are beginning to use Ray for scaling and other considerations. In a previous episode of this podcast, Hyperopt maintainer Max Pumperla also expressed admiration for Ray Tune.

Subscribe to our Newsletter:
We also publish a popular newsletter where we share highlights from recent episodes, trends in AI / machine learning / data, and a collection of recommendations.

Related content:

[Image: Soap Bubble from pxfuel]