How I Built the World’s Most Efficient Deepfake Detector with $100

Discover how to build a highly efficient deepfake detector with limited resources in this 39-minute Devoxx conference talk. Explore the world of ThisPersonDoesNotExist.com, an ultra-realistic face generator, and learn techniques to identify fake profile pictures with 100% reliability. Delve into the technical challenges of implementing attacks using TensorFlow models and customized ElasticSearch on limited hardware. Gain insights into basic machine learning concepts, Generative Adversarial Networks (GANs), and methods to generate nearly undetectable fake faces. Follow along as the speaker, Mathis Hammel, a tech evangelist and cybersecurity specialist, shares his expertise in programming competitions and Capture The Flag challenges. Covering topics from neural networks to database setup and image analysis, this beginner-friendly presentation offers valuable knowledge for combating disinformation and manipulation in the digital age.

Intro
About Coding Game
Demonstration
Introduction
Neural Network
Machine Learning Crash Course
Neural Networks
Neural Networks Examples
Gans
Generator
Experiment setup
Overfitting
Slow convergence
Computing power
Gun generated images
How to cheat
Another game
Indicators
Eyes
Script
Database
Downloader script
Resources
Images
Jetson Nano
FaceNet
Fuzzy search
Setup
Bonus points
When the image was generated
Demo time
Top 3 matches
Search speed
Continuity
Multiple variants
Questions