ProbabilitaML/pecman_l2d
Human-machine cooperation via learning-to-defer
What is learning-to-defer?
When making a prediction or decision, a machine learning model will consult a human expert if it is not confident of its prediction.
Example
- Self-driving cars: sometimes alert drivers to take control.
- Voice assistants (e.g., Siri): sometimes output "I'm not sure about that" instead of giving the wrong answers.
→ deferring helps to avoid mistakes and keeps things safe.
Why do we need human-machine cooperation?
<img src="img/motivation.png" alt="drawing" width="720"/>
Integrating human and machine into one system is to leverage:
- reliability of human experts, and
- efficiency of machine learning models.
→ highly-reliable and cost-effective systems.
Interactive game
https://huggingface.co/spaces/ProbabilitaML/pecman_l2d
<img src="img/qrcode.svg" alt="drawing" width="250"/>
What is pneumonia?
<img src="https://www.cell.com/cms/10.1016/j.cell.2018.02.010/asset/729eaec0-cf06-4872-b88c-2cadfbbf577f/main.assets/figs6_lrg.jpg" alt="pneumonia" />
The normal chest X-ray (left panel) depicts clear lungs without any areas of abnormal opacification in the image. Bacterial pneumonia (middle) typically exhibits a focal lobar consolidation, in this case in the right upper lobe (white arrows), whereas viral pneumonia (right) manifests with a more diffuse "interstitial" pattern in both lungs.
<br /> <img src="https://www.mdpi.com/diagnostics/diagnostics-10-00417/article_deploy/html/images/diagnostics-10-00417-g002.png" alt="pneumonia-example-2" />