Accelerating Deep Learning by Focusing on the Biggest Losers - in Tensorflow

Описание: Accelerating Deep Learning by Focusing on the Biggest Losers. Full code walkthough of Tensorflow Keras implemented two ways. #NeuralNetwork #TensorFlow #MachineLearning

[00:00] Intro
[01:00] Outline of notebook

[02:40] Beginning of code walk through
[03:20] MNIST import and helper functions
[05:00] 'Make Model' helper function
[06:00] Detailed description of neural net
[11:40] Model Shape
[14:30] Test helper function
[15:20] 'Regular' Training benchmark

[16:30] Worst K - my mini-batch implementation
[19:30] How worst K works

[22:15] Biggest Losers - as per the paper
[24:00] training function changes

[26:17] Result Graphs
[28:50] Review

[29:45] Unbalanced Classes
[34:00] Personal take

The original paper https://arxiv.org/abs/1910.00762

The colab notebook https://colab.research.google.com/dri...