Austin Powell

Austin Powell is a statistician turned data scientist using Python as an entry point into computer science. He tackles a diverse array of data and computational problems in his current position at Kaiser Permanente. There he evangelizes the use of Python for everything from scientific computing, to natural language processing to web scraping.

Speaker home page

Automated responses to questions about your health

Python & Libraries, AI & Data, Scale & Performance, Intermediate
8/18/2018 | 4:40 PM-5:10 PM | Fisher

Description

Have you ever searched online for some healthy issue? With Tensorflow models built from freely available data, and utilizing a domain specific ontology, learn how to bootstrapped a solution for automatically responding to the types questions you might ask on Google but need a more specific medical response.

Abstract

Have you ever searched online for some healthy issue? Chances are you gained little meaningful information from your search and perhaps made your doctor’s life that much harder when you went into their office with your newfound “knowledge”. With Tensorflow models built from freely available data, and utilizing a domain specific ontology, learn how to bootstrapped a solution for automatically responding to the types questions you might ask on Google but need a more specific medical response.

Using the open source Python library Tensorflow, I will go over the steps I took to create a model that generates medical answers to healthcare questions. It will discuss how to frame your machine learning project so that your time and efforts are rewarded with valuable insights that you can report to your team. Technical details such as model performance, choice of deep learning library and design as well as the basics of acquiring good data and cleaning it will be covered. Also covered will be healthcare industry specific challenges in natural language processing and what effort is necessary to create a question/answer system that produces meaningful results

More specifically it will touch on: - Entity extraction - Automated ontology generation - Building training sets locally and on cloud - Speaker recognition - Sentiment analysis - Speed comparisons between Tensorflow, Keras and PyTorch