1. Image Recognition with
Deep Learning using Apache
Spark and BigDL
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

2. Feed-forward network
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

3. 43
37
45
40
𝑦 = (𝑤𝑖 ∗ 𝑥𝑖)
?
?
?
?
?
?
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

4. -0.53
0.01
-0.17
0.70
0.51
?
?
𝑦 = (𝑤𝑖 ∗ 𝑥𝑖)
43
37
45
40
?
?
?
?
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

5. -0.53
0.01
-0.17
0.70
0.51
?
?
𝑦 = (𝑤𝑖 ∗ 𝑥𝑖)
43
37
45
40
-1.56
?
?
?
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

6. -0.53
0.01
-0.17
0.70
0.51
?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40
-1.56
?
?
?
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

7. -0.53
0.01
-0.17
0.70
0.51
?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40
?
?
?
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

8. 43
37
45
40
-0.12
0.13
0.21
-0.07
-0.05
?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
?
?
0
11.9
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

9. ?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40
-.11
?
0
11.9
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

10. ?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40 ?
0
11.9
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

11. ?
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40 0.15
0
11.9
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

12. -0.67
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40 0.15
0
11.9
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

13. 0
?
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40 0.15
0
11.9
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

14. 0
0.52
𝑦 = 𝑅𝑒𝐿𝑈( 𝑤𝑖 ∗ 𝑥𝑖 )
43
37
45
40 0.15
0
11.9
0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

15. Feed-forward network
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

16. Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

17. Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

18. Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

19. Fully Connected
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

20. Fully Connected
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

21. Fully Connected
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

22. Fully Connected
Input Size:
Connections:
40,000
1,600,000,000
200
200
200
200
10 layers: 16 billion
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

23. Convolutional Neural Networks
• LeNet-5 network developed in 1998 by Yann LeCun
• Torsten Hubel and David Wiesel
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

24. Hubel & Wiesel
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1363130/

25. Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

26. Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

27. Hierarchical Visual Cortex
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

28. Hierarchical Visual Cortex
Lines,
Dots
Orientation,
Movement
High-Level
Shapes
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

29. Local Receptive Fields
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1363130/

30. Convolutional Network
• Hierarchical processing
• Localized receptive fields
0
0
0
0
0
1
0
0
0
0
0
1
2
3
4
5
6
7
8
9
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

31. Fully Connected
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

32. Convolution
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

33. Convolution
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

34. Convolution
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

35. Convolution
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

36. Convolution
Only four weights
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

37. Convolution
0.10 -0.06
0.24 0.17
Filter
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

38. Source: https://www.amazon.com/Hands-Machine-Learning-Scikit-Learn-TensorFlow-ebook
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

39. Source: https://www.amazon.com/Hands-Machine-Learning-Scikit-Learn-TensorFlow-ebook
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
0 0 1 0 0
0 0 0 0 0
0 0 0 0 0
1 1 1 1 1
0 0 0 0 0
0 0 0 0 0
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

40. Pooling
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/Source: https://cs231n.github.io/convolutional-networks/

41. Convolutional Network
• Hierarchical processing
• Localized receptive fields
0
0
0
0
0
1
0
0
0
0
0
1
2
3
4
5
6
7
8
9
Convolution Pooling Convolution PoolingInput FC FC
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

42. Convolutional Network
0
0
0
0
1
0
0
0
0
0
0
1
2
3
4
5
6
7
8
9
Convolution Pooling Convolution PoolingInput FC FC
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/Source: https://www.clarifai.com/technology

43. Workshop
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/

44. Questions?
•GitHub: https://github.com/alex-kalinin/lenet-bigdl
•LinkedIn: https://www.linkedin.com/in/alexkalinin/
Alex Kalinin
https://www.linkedin.com/in/alexkalinin/