Networks are like onions: Practical Deep Learning with TensorFlow
summary
Starts with a one-line TensorFlow model that flattens 28x28 MNIST pixels to a 784→10 dense layer and gets ~83% accuracy, then adds a 1024-unit hidden layer with ReLU (using tf.layers.dense) to reach ~94%. Replaces dense hidden layers with conv blocks: 32 filters 5x5 + 2x2 max-pool, then 64 filters 5x5 + pool, flatten to FC and hit ~97%. Training uses batches, softmax cross-entropy loss, optimizer.minimize, and applies dropout 0.5 during training.
AI-written summary · about
chapters (7)
- 0:00Start
- 6:25Neural Networks
- 14:16What has changed?
- 29:48Hidden layer
- 36:18Convolutional Neural Network
- 38:34Pooling
- 52:53Dropout
related talks
01h 06mDeep Learning with TensorFlow
49mDeep Learning With Python & Tensorflow
27mDemystifying Deep Neural Networks
54mNeural networks by example
45mThe most soft and fluffy intro into Machine Learning and Deep Neural Networks
56mTutorial on Deep Learning II
35mDeconstructing Deep Learning
01h 07mDeep Learning with TensorFlow