~/talk

Deconstructing Deep Learning

Talkat Big Data Conference Europe 2019 · 2019-11-26..11-28· [machine-learning]
slides:none on file

Covers concrete implementations: a CNN built in Keras/TensorFlow runnable in Google Colab, with five hidden layers, 32 convolutional filters, max pooling and 25% dropout, trained in batches of 256 to about 90% accuracy on Fashion‑MNIST; an LSTM with 32‑dim word embeddings, trained three epochs on IMDB reviews, reporting ~87% accuracy. Explains backprop/gradient descent and shows feature maps, misclassified examples, and practical preprocessing steps that changed results.

AI-written summary · about

  1. 0:00Start
  2. 0:26Agenda
  3. 1:18Deep Learning Biologically inspired machine learning with multi
  4. 2:46Artificial Neural Network Structur
  5. 3:51Deep Artificial Neural Network
  6. 4:10Artificial Neural Network Node
  7. 5:12Activation Functions
  8. 7:00Training an Artificial Neural Netwo
  9. 9:05Updating Weights via Gradient De
  10. 10:22Deep Learning 101 Summary
  11. 11:05How computers «see images
  12. 13:25Another Approach: Convolutional Fi
  13. 14:18Applying Convolutional Filters
  14. 15:34Applying ReLU Activation Functio
  15. 16:02Dimensionality Reduction via Pooli
  16. 16:58Feature Identification and Dimensionality
  17. 17:37Use Case: Fashion MNIST
  18. 18:03CNN Implementation
  19. 18:44CNN Architecture Overview
  20. 22:14Training and Testing the CNN
  21. 22:39CNN Test Results
  22. 23:09Some examples of incorrectly classified
  23. 23:33CNN Summary
  24. 24:46Recurrent Neural Networks to the Re
  25. 25:15Unrolling a Recurrent Neural Netw
  26. 25:54Long Short-Term Memory (LSTM) Net
  27. 26:11A LSTM Node
  28. 26:48Use Case : IMDB Reviews
  29. 27:21Example Positive Review
  30. 27:35Pre-Processing Dataset
  31. 28:12Word Embedding for NLP
  32. 29:00LSTM Implementation
  33. 30:45Training and Testing the LSTM
  34. 31:22LSTM Test Results
  35. 32:55Example of an incorrectly classified re
~/search
type to search 6,249 conferences, 138,987 speakers and 185,172 talks
~/help
cmdctrl k
search every conference, speaker and series
/
focus the filter search
h
home
c
conferences
s
speakers
e
series
t
light / dark — the status bar button does the same
?
this help
esc
close this, then clear every filter — or just the search, while you are typing in it

Ignored while you are typing. Pages without a filter rail have no search to focus.