TensorFlow and Deep Learning without a PhD, Part 1

on YouTube ↗56m

" summary

Builds an MNIST digit recognizer in TensorFlow that moves from a single-layer softmax baseline (92% accuracy) to a deep model with relu activations, mini-batches, exponential learning-rate decay and gradient descent, then three convolutional layers (e.g. 5x5 and strides to 7x7), a fully connected layer with dropout on the fully connected layer, and tuning of filter counts and batch sizes to reach 99.3% accuracy. Uses TensorFlow primitives, convolutional layers with 5x5 filters.

AI-written summary · about

" chapters (11)

1. 0:00Intro
2. 0:49Hello World: handwritten digits classification - MNIST
3. 13:48TensorFlow - initialisation
4. 23:27TensorFlow - full python code
5. 25:42Let's try 5 fully-connected layers!
6. 30:01Demo - noisy accuracy curve ?
7. 31:21Learning rate decay
8. 35:18All the party tricks
9. 37:12Overfitting
10. 40:29Convolutional layer
11. 45:06Convolutional neural network

" related talks

• 43mTensorFlow and Deep Learning without a PhDLucio Floretta · Codemotion Milan 2017
• 01h 06mDeep Learning with TensorFlowSeth Juarez · NDC Sydney 2018
• 01h 07mDeep Learning with TensorFlowSeth Juarez · NDC Oslo 2018
• 34mML from a software engineer's perspectiveMarijn van Zelst · Codemotion Amsterdam 2017
• 01h 04mWhat you always wanted to know about Deep Learning, but were afraid to askWei-Meng Lee · NDC London 2020
• 57mTensorflow, deep learning and modern convolutional neural nets, without a PhDMartin Gorner · Devoxx Belgium 2017
• 01h 29mHands-on introduction to Deep Learning with Keras and TensorflowRodrigo Agundez · PyData Amsterdam 2018
• 49mDeep Learning With Python & TensorflowIan Lewis · PyCon Singapore 2016