Energy based models II
This lecture is a foundationational lecture for the concept of energy based models with a particular focus on the joint embedding method and latent variable energy based models 8LV-EBMs) and is a part of 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: Introduction to Deep Learning, Parameter sharing, and Introduction to Data Science or a Graduate Level Machine Learning.
Chapters:
00:00:00 – Welcome to class
00:00:17 – Training of an EBM
00:04:27 – Contrastive vs. regularised / architectural methods
00:05:21 – General margin loss
00:09:34 – List of loss functions
00:13:45 – Generalised additive margin loss
00:17:53 – Joint embedding architectures
00:21:29 – Wav2Vec 2.0
00:27:14 – XLSR: multilingual speech recognition
00:29:16 – Generative adversarial networks (GANs)
00:37:24 – Mode collapse
00:41:45 – Non-contrastive methods
00:43:19 – BYOL: bootstrap your own latent
00:44:27 – SwAV
00:46:45 – Barlow twins
00:51:29 – SEER
00:54:29 – Latent variable models in practice
00:57:34 – DETR
01:01:21 – Structured prediction
01:04:53 – Factor graph
01:12:47 – Viterbi algorithm whiteboard time
01:30:24 – Graph transformer networks
01:46:48 – Graph composition, transducers
01:48:38 – Final remarks