Quality Control to avoid GIGO in Deep Learning Models
summary
A Python pipeline computes dozens of image metrics and scores each image with an IsolationForest (score range −1 to 0) then an interactive Shiny UI lets scientists slide the cutoff and immediately see which images are flagged so a plate of ~3,000–13,000 images avoids the ~10% automatic-reject rate that would throw away valid data. Deployed via AWS Connect, the web app (first version built in ~2 months) keeps a human in the loop and yields cleaner training sets for downstream deep learning.
AI-written summary · about
chapters (10)
- 0:00Avoiding GIGO in deep learning
- 1:10Quality control in cell imaging
- 2:36Cell painting and screening
- 4:25Image QC in pipeline
- 5:31Automating QC with Python
- 8:03Shiny app for thresholding
- 10:27Human-in-the-loop validation
- 11:59Training with clean data
- 13:23Team collaboration
- 13:44Q&A
related talks
34mHalf Hour of Flower: Planting the seeds for great Image Quality
18mDeep learning for Space imagery applications
40mHow to Implement a Production Ready Deep Learning Model in 12 Weeks
01h 01mA practical guide to deep learning
38mBuilding robust machine learning systems
29mLabeling Tools are Great, but What About Quality Checks?
30mHow to Properly Test ML Models & Data
28mData Curation: Transforming Bytes into AI Gold