Deep Learning Course (NYU, Spring 2020) - Week 14 – Lecture: Structured prediction with energy based models

Course website: http://bit.ly/pDL-home Playlist: http://bit.ly/pDL-YouTube Speaker: Yann LeCun Week 14: http://bit.ly/pDL-en-14 0:00:00 – Week 14 – Lecture LECTURE Part A: http://bit.ly/pDL-en-14-1 In this section, we discussed the structured prediction. We first introduced the Energy-Based factor graph and efficient inference for it. Then we gave some examples for simple Energy-Based factor graphs with “shallow” factors. Finally, we discussed the Graph Transformer Net. 0:00:25 – Structured Prediction, Energy based factor graphs, Sequence Labeling 0:18:06 – Efficient Inference for Energy-Based Factor Graph and Some Simple Energy-Based Factor Graphs 0:43:30 – Graph Transformer Net LECTURE Part B: http://bit.ly/pDL-en-14-2 The second leg of the lecture further discusses the application of graphical model methods to energy-based models. After spending some time comparing different loss functions, we discuss the application of the Viterbi algorithm and forward algorithm to graphical transformer networks. We then transition to discussing the Lagrangian formulation of backpropagation and then variational inference for energy-based models. 1:00:22 – Comparing Losses and the start of language models as graphs 1:15:18 – Forward algorithm in Graph Transformer Networks 1:32:53 – Lagrangian formulation of back prop and neural ODE 1:48:42 – Variational Inference in terms of Energy