Practical Deep Learning for Data Scientists

on YouTube ↗02h 26m

" summary

Hands-on workshop that builds an animal classifier using PyTorch on GPU using Domino and the Snapshot Serengeti images, shrinking classes to four and sampling ~200 images per class, swapping in a pre‑trained ResNet50 and freezing then fine‑tuning layers. Practical wins: image caching and pinned memory cut preprocessing from ~0.05s to ~0.008s per image (roughly 10x speedup), epoch time drops from ~1m50s to ~26s for five epochs, and tuned runs exceed 90% validation accuracy, so the model trains and generalizes.

AI-written summary · about

" chapters (59)

1. 0:00Start
2. 2:05Introduction
3. 2:48When Should You Use Um Deep Learning
4. 3:27General Examples of Deep Learning
5. 5:01What a Neural Network Is
6. 5:14Review
7. 6:32Hidden Layers
8. 8:32Weights
9. 9:22Sigmoid Function
10. 10:51Neural Network Learning
11. 13:13Cost Function
12. 14:08Back Propagation
13. 16:23Stochastic Gradient Descent
14. 17:30Deep Convolutional Network
15. 18:26Convolutional Neural Network
16. 19:05Feature Map
17. 19:31Recurrent Neural Networks
18. 23:40Fork a Project
19. 25:15Start Workspaces
20. 26:40Forking
21. 27:54Deep Learning in Python
22. 28:59Basics of Pytorch
23. 29:56Data Loaders and Transforms
24. 30:29Transforms
25. 45:09Review of Data Cleaning
26. 45:34Create a Second Workspace
27. 46:11Overview
28. 48:09Overview of the Data
29. 53:56Workspace
30. 55:30Loading Data into Pi Torch
31. 56:38Data Loader
32. 1:09:28Deep Learning and Practice
33. 1:12:20Cnn Architecture
34. 1:12:52Convolutional Layer
35. 1:14:59Activation Functions
36. 1:15:50Softmax
37. 1:16:22Pooling Layer
38. 1:16:32Max Pooling
39. 1:16:44Average Pooling
40. 1:19:36Transfer Learning
41. 1:21:43Reproducibility
42. 1:23:32Training a Model
43. 1:25:49What Loss Function Do You Use
44. 1:26:08Loss Functions
45. 1:27:22Relative Entropy
46. 1:31:13Data Set
47. 1:34:22Train Test Split
48. 1:37:45Set Up a Loader
49. 1:39:19Model Training Loop in Pi Torch
50. 1:40:43Train Phase and a Validation Phase
51. 1:44:05Transfer Learning
52. 1:46:30Convolution Layer
53. 1:49:11Cache Item
54. 1:50:38Cache Transform
55. 1:52:39Pinning Memory
56. 1:57:44Optimizing Model Performance
57. 1:58:16Epochs
58. 1:58:49Optimizers
59. 1:59:34Learning Rates

" related talks

• 32mYou Can Do Deep Learning!William Horton · PyOhio 2018
• 01h 01mA practical guide to deep learningTess Ferrandez-Norlander · NDC London 2019
• 01h 27mHands-On Computer Vision with PyTorchRobert Alvarez · PyData NYC 2022
• 53mDeep Learning in the world of little poniesGaliya Warrier · NDC Oslo 2019
• 01h 02mDeep Learning in the world of little poniesGaliya Warrier · NDC Sydney 2018
• 59mOpening KeynoteKatharine Jarmul · PyCon APAC 2018
• 01h 32mIntroduction to PyTorchRobert Alvarez · PyData Miami 2019
• 01h 00mA gentle introduction to Deep LearningLaurent Bugnion · NDC Porto 2022