
Lecture 8 of 33 · Yann LeCun’s Deep Learning
NYU Data Science
Yann LeCun’s Deep Learning · 33 lectures
33 lectures
History and resources
L - Gradient descent and the backpropagation algorithm
Neural nets: rotation and squashing
L - Modules and architectures
Tools, classification with neural nets, PyTorch implementation
L - Parameter sharing: recurrent and convolutional nets
L - ConvNet in practice
1 - Natural signals properties and the convolution
2 - Recurrent neural networks, vanilla and gated (LSTM)
L - Joint embedding method and latent variable energy based
1 - Latent Variable Energy Based Models (LV-EBMs), inference
2 - But what are these EBMs used for?
L - Latent variable EBMs for structured prediction
Latent Variable Energy Based Models (LV-EBMs), training
L - PCA, AE, K-means, Gaussian mixture model, sparse coding
Unsupervised learning: autoencoding the targets
L - Self-supervised learning and variational inference
From LV-EBM to target prop to (vanilla, denoising, contractive
L - Differentiable associative memories, attention
AE, DAE, and VAE with PyTorch
L - Self-supervised learning in computer vision
Self / cross, hard / soft attention and the Transformer
L - Speech recognition and Graph Transformer Networks
Graph Convolutional Networks (GCNs)
L - Low resource machine translation
Planning and control
L - Optimisation for Deep Learning
The Truck Backer-Upper
L - Lagrangian backpropagation, final project winners
Prediction and Planning Under Uncertainty
AI2S Xmas Seminar - Dr. Alfredo Canziani
P - Contrastive joint embedding methods (JEMs)
P - Non-contrastive joint embedding methods (JEMs)