Machine learning is as valuable as the problems it solves. With this introductory talk in cooperation with AI Innovation Center and Ipsumio, we aim to provide the power of this tool to professionals dealing with important problems in healthcare, physics, manufacturing, and many others. This is a non-technical talk for domain experts to get a high-level understanding of this technology.
5. Symbolic AI
Machine Learning
Supervised
Unsupervised
Symbolic AI
Machine Learning
Supervised
Unsupervised
Artificial Intelligence
• Neural Network
• Linear Regression
• Random Forest
Deep Learning
5
Regression Classification
Particle size measurement Cell A or Cell B
35. Introduction
Data – Learning Algorithm – Model
Code
Overfitting/Underfitting
ML systems
How to approach ML projects
36. 36
How to find an ML project
Image processing
• Classify images
• Localize objects inside images
• Count objects in video
• Measure x
Feature 1 Feature 2 Target
Example 1
Example 2
Example 3
Example 4
…
Sensor
Raw
data
Measurement
Non-ML
Model
Optimization
38. Are you solving the correct problem?
Do you have the signal inside your dataset?
Does your sample represent the population?
Do you have non-random noise in your dataset?
Surpass a physician’s ability to
predict disease
Predict normal heart
function
Body temperature
39. 39
Class activation maps
Cardiomegaly prediction
https://medium.com/@jrzech/what-are-radiological-deep-learning-models-actually-learning-f97a546c5b98
40. Positive/Negative source
"We show that
similar results can be obtained
using X-ray images
that do not contain most of the lungs."
https://arxiv.org/abs/2004.12823
40
Negative data source COVID-19 data source
A, B, C … X, Y, Z …
415 studies
0 clinically useful
46. External testing
Hospital X
Dr. A
ProspecOve
Dr. B Dr. C Dr. D
Hospital Z
Country K
Hospital X Hospital X Hospital Y
Country K Country K Country K Country L
Train/Test
Dr. A
46
47. Training
Dr. A Dr. B Dr. C Dr. D
Hospital Z
Hospital X Hospital X Hospital Y
Country K Country K Country K Country L
47