Deep Learning · MT2 · index

Questions and complete answers

  1. Model answer: explain an artificial neuron
  2. Model answer: why a network needs nonlinear activation
  3. Model answer: forward propagation, backpropagation and updating
  4. Model answer: batch, stochastic and mini-batch training
  5. Model answer: momentum and Nesterov
  6. Model answer: AdaGrad, RMSProp and Adam
  7. Model answer: regularization and the L1–L2 difference
  8. Model answer: generalization and early stopping
  9. Model answer: dropout with training and inference explained
  10. Model answer: CNN structure and parameter sharing
  11. Model answer: shape, activations and parameters
  12. Model answer: feature extraction and fine-tuning
  13. Model answer: AlexNet, Inception and ResNet
  14. MT1A Q2
  15. MT1A Q3
  16. MT1A Q4
  17. MT1B Q3
  18. MT1B Q4
  19. F25 Q1c
  20. F25 Q2b
  21. F25 Q3b
  22. F25 Q4a
  23. F25 Q5b
  24. F25 Q6a
  25. F25 Q7b
  26. R26 Q1c
  27. R26 Q2b
  28. R26 Q3b
  29. R26 Q4a
  30. R26 Q5b
  31. R26 Q6a
  32. R26 Q7b
  33. F23 Q1b
  34. F23 Q1d
  35. F23 Q3b
  36. F23 Q4a
  37. F23 Q5a
  38. F23 Q6b
  39. F23 Q7b
  40. R24 Q1c
  41. R24 Q1d
  42. R24 Q2b
  43. R24 Q3b
  44. R24 Q4b
  45. R24 Q5a
  46. R24 Q6b
  47. F22 Q1a
  48. F22 Q1b
  49. F22 Q3a
  50. F22 Q3b
  51. F22 Q4a
  52. F22 Q4b
  53. F22 Q5b
  54. F22 Q6a
  55. F22 Q6b
  56. R23 Q1b
  57. R23 Q1d
  58. R23 Q2a
  59. R23 Q2b
  60. R23 Q4a
  61. R23 Q4b
  62. R23 Q6a
  63. R23 Q6b
  64. R23 Q7b
  65. Drill 1 Signs and brackets
  66. Drill 2 Half sum versus mean
  67. Drill 3 Natural log and probability
  68. Drill 4 Dot product and bias
  69. Drill 5 Matrix rows receive inputs
  70. Drill 6 Chain rule on a tiny loss
  71. Drill 7 Sigmoid with binary cross entropy
  72. Drill 8 Activation contrast
  73. Drill 9 Softmax and true class loss
  74. Drill 10 Sigmoid neuron gradient
  75. Drill 11 Full ReLU backpropagation
  76. Drill 12 Two output sigmoid backpropagation
  77. Drill 13 A negative ReLU blocks this path
  78. Drill 14 Batch and sequential SGD
  79. Drill 15 Incomplete mini batch
  80. Drill 16 GD variation 1
  81. Drill 17 GD variation 2
  82. Drill 18 Momentum variation 1
  83. Drill 19 Momentum variation 2
  84. Drill 20 Nesterov variation 1
  85. Drill 21 Nesterov variation 2
  86. Drill 22 AdaGrad variation 1
  87. Drill 23 AdaGrad variation 2
  88. Drill 24 RMSProp variation 1
  89. Drill 25 RMSProp variation 2
  90. Drill 26 Adam variation 1
  91. Drill 27 Adam variation 2
  92. Drill 28 Regularized objective
  93. Drill 29 L1 update on positive and negative weights
  94. Drill 30 L2 update
  95. Drill 31 Inverted dropout
  96. Drill 32 Convolution with a negative coefficient
  97. Drill 33 Padding stride and channels
  98. Drill 34 Max and average pooling
  99. Drill 35 Complete small CNN
  100. Drill 36 Reverse a spatial formula
  101. Drill 37 Bottleneck cost with new values
  102. Drill 38 Residual addition
  103. Drill 40 Mixed convolution pool dense and loss
  104. Drill 41 Mixed backpropagation and L2
  105. Drill 42 Mixed dropout and a momentum update
  106. Drill 43 Mixed CNN dimensions softmax and loss
  107. Drill 44 Inception concatenation and parameter count
  108. Drill 45 Projection shortcut