Introduction to Deep Learning and Neural Networks.

Transcript

1. Introduction to Deep Learning and Neural Networks. Bedanta Bikash Borah @iamBedant
4. Agenda 1. What is Deep Learning? 2. Why Deep Learning is taking off? 3. How Deep Learning works? 4. Training. 5. Example MNIST. 6. Code Sample. 7. Few extra concepts.
5. Deep Learning The term Deep Learning refers to training very large Neural Network
6. What is Neural Network? House Size (X) Price (Y) 500 5 Laks 600 8 Laks 700 9 Laks 800 13 Laks 900 12 Laks 1100 18 Laks 1200 20 Laks
7. Housing Price Prediction 0 laks 5 laks 10 laks 15 laks 20 laks 000 sq ft 300 sq ft 600 sq ft 900 sq ft 1200 sq ft What is Neural Network?
8. What is Neural Network? x Y Size of the House Price of the House Neuron Function F Input Output
9. What is Neural Network? Size No of bedrooms Locality Society Y Family Size Walkability School/Market Quality
10. What is Neural Network? X1 X2 X3 X4 Y Size No of bedrooms Locality Society
11. Deep Neural Network
12. Why DeepLearning taking off? Deep Neural Network Medium Neural Network Traditional ML Algorithm Data Performance Not well defined
13. Why DeepLearning taking off? 1. Data 2. Computation 3. Algorithm
14. How DeepLearning Works? X1 X2 X3 X4 Y
15. How DeepLearning Works? X1 X2 X3 X4 Y f W1 W2 W3 W4 X1* W1 + X2 * W2 + X3 * W3 +X4 * W4 Z f( )= Relu (x) or Sigmoid(x) *ignoring bias for simplification Z =
16. How DeepLearning Works?
17. How DeepLearning Works? X1 X2 X3 X4 Y
18. Interviewer: What is your biggest strength? Me: I am an expert in machine learning. Interviewer: What’s 9 + 10? Me: It’s 3. Interviewer: Nowhere near. It’s 19. Me: It’s 16. Interviewer: Wrong. It’s still 19 Me: It’s 18. Interviewer: No. It’s 19 Me: It’s 19. Interviewer: You’re hired
19. Training X1 X2 X3 X4 Y Y’ Random Initialisation
20. Training 1.Quadratic cost 2.Cross-entropy cost 3.Exponential cost Cost Function:
21. Training Grad Gradient Descent
22. Training Learning Rate (alpha)
23. Training Large Learning rate
24. Training Learning Rate (alpha)
25. Training New Weights = Existing Weights Learning Rate - * Gradient ( )
26. Interviewer: What is your biggest strength? Me: I am an expert in machine learning. Interviewer: What’s 9 + 10? Me: It’s 3. Interviewer: Nowhere near. It’s 19. Me: It’s 16. Interviewer: Wrong. It’s still 19 Me: It’s 18. Interviewer: No. It’s 19 Me: It’s 19. Interviewer: You’re hired.
27. Example MNIST
28. MNIST 60,000 training samples 10,000 test samples
29. MNIST
30. MNIST = 28 x 28 = 784
31. MNIST Softmax Image Vector Neural Network Layers
32. MNIST Softmax Image Vector Neural Network Layers
33. MNIST Softmax Image Vector Neural Network Layers
34. MNIST Softmax Image Vector Neural Network Layers
35. MNIST Softmax Image Vector Neural Network Layers
36. MNIST **from three blue one brown’s “But, what is a neural network?” video**
37. Talk is cheap show me the code.
38. Advanced MNIST CNN (convolutional neural network)
39. Interviewer: What is your biggest strength? Me: I am an expert in machine learning. Interviewer: What’s 9 + 10? Me: It’s 3. Interviewer: Nowhere near. It’s 19. Me: It’s 16. Interviewer: Wrong. It’s still 19 Me: It’s 18. Interviewer: No. It’s 19 Me: It’s 19. Interviewer: You’re hired.
40. Interviewer: What is your biggest strength? Me: I am an expert in machine learning. Interviewer: What’s 9 + 10? Me: It’s 3. Interviewer: Nowhere near. It’s 19. Me: It’s 16. Interviewer: Wrong. It’s still 19 Me: It’s 18. Interviewer: No. It’s 19 Me: It’s 19. Interviewer: What’s 20 + 10? Me: It’s 19
41. Advanced MNIST Overfitting (Regularization, Dropout)
42. Extras
43. Reference https://github.com/iamBedant/CMRIT-Deeplearning-TechTalk-Demo Simple MNIST Example https://github.com/iamBedant/TensoreFlowLite Android TFLite Example Others: https://www.tensorflow.org/ https://keras.io/
44. Thank You !!! @iamBedant