Taming the Machine (feat. Nicolas Cage)

Laborde connects machine learning to React Native’s “learn once, write anywhere” approach. He uses spell-checking as an example: one model can work across languages when trained on different datasets, instead of building a separate solution for each language. He focuses on supervised learning and emphasizes that its underlying math is approachable.

Laborde describes an image as a set of numerical inputs, such as the 784 grayscale values in a 28-by-28 image. He shows how iOS Vision can identify faces and landmarks, and introduces React Native Vision as a way to use these capabilities from React Native. In a live demo, a downloaded food model classifies hot dogs, fries, pancakes, and hamburgers.

Laborde changes the demo from food classification to face detection by swapping the model and provider. The updated app draws a detector over faces and checks whether someone resembles the person associated with the Declaration of Independence gag. The person flagged in the demo is Matt Hargett, not Nicolas Cage.

Laborde explains that he gathered Nicolas Cage photos, collected other celebrities as examples, removed faces from images, and checked the data for errors. Early training did not work: accuracy and loss failed to diverge, so he combined his facial data with a researcher’s model trained on 500,000 images. He then quantized the large model and demonstrated it identifying Cage, including a poor-quality photo at 70% confidence while identifying his friend Frank as not Cage at 99%.