Unsupervised learning: autoencoding the targets
This lecture covers advanced concept of energy based models. The lecture is a part of the Advanced energy based models modules of the the Deep Learning Course at CDS, a course that covered the latest techniques in deep learning and representation learning, focusing on supervised and unsupervised deep learning, embedding methods, metric learning, convolutional and recurrent nets, with applications to computer vision, natural language understanding, and speech recognition. Prerequisites for this course include: Energy based models I, Energy based models II, Energy based models III, and Introduction to Data Science or a Graduate Level Machine Learning.
Chapters:
00:00 – 2021 edition disclaimer
02:28 – Unsupervised learning and generative models
05:42 – Input space interpolation
08:24 – Latent space interpolation
10:54 – Conditional generative networks
13:37 – Style transfer
16:21 – Super resolution
21:37 – Inpainting
23:19 – Caption to image (Dall-e)
28:24 – Definitions: x, y, z
31:27 – Recap: conditional latent variable EBM
32:12 – Recap: energy function
32:39 – Softmin training recap → autoencoder via amortised inference
38:40 – Reconstruction energies
39:24 – Loss functional
43:59 – Under and over complete hidden layer
48:58 – Denoising autoencoder
54:28 – Nearest neighbourhood denoising autoencoder
55:00 – Sparse autoencoder
55:20 – Final remarks
