This talk was given at Abstractions Pittsburgh on August 18, 2016. It is an introduction to the basics of machine learning and how they can be applied to healthcare.
How is Real-Time Analytics Different from Traditional OLAP?
Abstractions Conference 2016 - Machine Learning in Healthcare – ML for the Rest of Us
1. Machine Learning in Healthcare – ML for the Rest of Us
Mohinder Dick
Senior Software
UPMC Enterprises
August 18, 2016
Email: dickm@upmc.edu
Twitter: @mobyware
2. |
Goal
Breakdown the perception that machine learning is too complex.
Address assumptions of what machine learning IS, and what it’s NOT.
Impart the basics of machine learning.
Goal is to provide a soft landing for a topic that can be the stuff of science fiction
Re demonstrate, illustration in healthcare
If time permits we will do so with a quick demo.
Machine learning is complex and vast:
Research
Terms
Technologies
Consider two visions of ML, one is way to complex.
This is good to aspire to:
ANN
Threat modelling
Advance image recognition
Arrow anecdote
Vision 2: Approachable for most of us here.
Use simpler types of machine learning to serve clinical context:
Facial recognition: OpenCV to find face. Glasses
Estimate length of stay in hospital
Determine which drugs are likely to be prescribed
Both features and targets can be categorical or continuous.
Prediction is the value of the target variable the machine comes up with.
Label is the value you give it during training.
Linear models use weights for each feature.
Classification is based on regions. Evaluate by percentage of TP, FP, TN, FN
Linear regression is based on the delta in the predicted value. Use measures like the RMSE
Haars are rectangular regions
Models for facial detection are typically non-linear. Trees or Net (neural nets)
Haar cascades – Viola & Jones (real-time image detection)
Both features and targets can be categorical or continuous.
All free and/or open source
eBay’s tech blog.
Spark application used to cluster sellers on eBay and store in HDFS