Link to the video: https://youtu.be/hNuhdf-fL_g
This is a submission to Veritasium's Sci-Comm contest. The target audience for this is the general audience and beginners to deep learning, so we tried our best to distill the main points of how neural nets are trained and how this relates to adversarial attacks into 1 minute.
We also made Google Colab notebooks with both an attack demo (using EfficientNet and Goodfellow et al.'s Fast Gradient Sign Method for attacking), and the Manim Community animation code. Manim is a cool math animation library originally created by 3Blue1Brown. We used it to make some of the animations in this video.
Interestingly, this contest originated from an UCLA physics professor losing a $10k bet to Veritasium, who used this money to create the contest. Every view will help us in the contest!
Top comments (0)