Unsupervised Learning with Autoencoders

Explore the world of unsupervised learning through autoencoders in this 49-minute conference talk by Christoph Henkelmann at MLCon. Dive into the basic concept of autoencoders, various architectural variants, and their diverse applications. Learn about CNN autoencoders, the importance of the bottleneck layer, and practical use cases including anomaly detection, denoising, and similarity detection. Discover the power of generative autoencoders, intrinsic space exploration, and the advantages of variational autoencoders (VAE). Gain insights into real-world applications of unsupervised learning techniques in industry, and leave with a comprehensive understanding of this powerful machine learning approach.

Intro
UNSUPERVISED TRAINING
CNN AUTOENCODERS
THE BOTTLENECK
EVEN MORE EXAMPLES
ANOMALY DETECTION
USE CASE 1: ANOMALY EXAMPLES
DENOISING
PRETRAINING
SIMILARITY DETECTION
GENERATIVE AUTOENCODERS
INTRINSIC SPACE & DIMENSION
PAC MAN'S INTRINSIC SPACE
THE IDEAL PAC-MAN BOTTLENECK
BACK IN REALITY...
THE VARIATIONAL AUTOENCODER
ADVANTAGES OF THE VAE
UNSUPERVISED LEARNING AT DIVISIO
SUMMARY COMING UP IN OUR BLOG