Training machine learning models to de identify sensitive user data (part 1)

Join Bugout dev and data science community https://join.slack.com/t/bugout-dev/shared_invite/zt-fhepyt87-5XcJLy0iu702SO_hMFKNhQ

Teams that maintain APIs, command line tools, and libraries use Bugout to collect crash reports from their users. Crash reports often contain sensitive, identifying information about the end user.

We care about preserving the end user's privacy to the maximum possible extent.

We are kicking off a series of live coding streams in which @zomglings aka Neeraj Kashyap, CEO of Bugout, will create machine learning models which deidentify crash reports.

These streams will cover:

the collection of training data
the definition and training of the actual models
model deployment (to our servers as well as our clients)
Here’s our rough plan:

First stream on May 19th at 9am PT: We will collect a dataset of crash reports by crawling GitHub.

Second stream (TBD): We will discuss possible ML model definitions for crash report deidentification and establish a baseline for their performance.

Third stream (TBD) – We will set up experiments to compare the different models. We will kick off these experiments at the end of the stream.

Fourth stream (TBD) - We will review the results of our experiments, decide on a model architecture to move into production, and create a deployment pipeline.

Neeraj holds a PhD in math from Indiana University. He has created algorithms to diagnose neurodegenerative diseases. Before creating Bugout, Neeraj has worked at Google on TensorFlow and Cloud AI teams.

If you have questions or want to be a part of our growing developer community: https://join.slack.com/t/bugout-dev/shared_invite/zt-fhepyt87-5XcJLy0iu702SO_hMFKNhQ