Crossword

1234567891011121314151617181920212223242526272829303132333435363738394041424344
Across
  1. 4. Regularization technique, randomly omitting units.
  2. 5. LSTM gate storing important info.
  3. 9. Learning based on connectedness.
  4. 11. Activation for multi-class classification.
  5. 13. Layers in between input and output.
  6. 14. Activation point for a neuron.
  7. 15. Learning with labeled training data.
  8. 16. Model's sensitivity to training data.
  9. 20. Loss function for classification.
  10. 22. Function defined by different equations.
  11. 23. Distribution characterized by a mean and variance.
  12. 24. Derivative indicating slope direction.
  13. 27. Accuracy of positive predictions.
  14. 28. Convolutional kernel in CNNs.
  15. 31. Basic unit of a neural network.
  16. 35. Small matrix for convolutional operations.
  17. 36. Training algorithm adjusting weights.
  18. 39. Reshaped into a one-dimensional array.
  19. 40. Optimization inspired by natural selection.
  20. 42. Large dataset for image classification.
  21. 43. Adaptive Linear Neuron.
  22. 44. Activation function, outputs between 0 and 1.
Down
  1. 1. Model configuration setting.
  2. 2. Technique preventing overfitting.
  3. 3. Connection point between neurons.
  4. 6. Simplest form of a neural network.
  5. 7. Subset of the training data.
  6. 8. Predicting numerical values.
  7. 10. Optimization algorithm enhancement.
  8. 11. Random or probabilistic.
  9. 12. Step size in convolutional operations.
  10. 17. Multi-branch convolutional architecture.
  11. 18. LSTM gate discarding unnecessary info.
  12. 19. Additional parameter, helps fit data.
  13. 21. Basic or standard version.
  14. 25. Rectified Linear Unit activation.
  15. 26. Adjusting pre-trained model for new task.
  16. 29. Dataset used to tune hyperparameters.
  17. 30. Activation outputs -1 or 1.
  18. 32. One pass through the entire training dataset.
  19. 33. Model fits training data too closely.
  20. 34. Information flows in one direction.
  21. 37. Adding extra pixels to input data.
  22. 38. Pioneering deep CNN architecture.
  23. 41. Down-sampling technique in CNNs.