# deepsense.ai interview - Michał Zobniów 2021

## Im więcej * tym mniej ważne

### Pure ML
1. Supervised Learning - Unsupervised Learning
2. K-nearest neighbors
3. Linear Regression
5. High Bias - High Variance
6. Regularization
7. Ridge Regression
8. Quantile regression***
9. Parameters - Hyperparameters
10. Statistical inference***
11. Naive Bayes Classifier
12. Logistic Regression
14. Logistic Regression for multiclass classification
15. Huber - pseudo-Huber Loss*
16. A General and Adaptive Robust Loss Function***
17. Feature Selection
* Scoring Based Methods
* Wrapper Methods
* LASSO
* LARS**
18. Decision Trees
* Purity Criterions
* Categorical - Numerical splits
* Dealing with Missing values
* DT for regression
* Pruning
19. Classifier Bagging
20. Random Forest
21. Boosting
22. AdaBoost
23. XGBoost
25. Kernels
26. SVM
27. Simplified PAC Theory****
28. K-means
29. Online K-means
30. Kohonen Maps****
31. Gaussian Mixture Models and EM****
32. Principal component analysis
33. Probabilistic Graphical Models****
34. Unbalanced data
### NN
1. Gradient descent
2. Batch, Mini Batch & Stochastic Gradient Descent
3. Activation functions
4. SGD
5. R-Prop and RMSProp
6. ADAM
7. NN Weight initialization
8. Early Stopping
9. Polyak weigth averaging*
10. Data Augmentation
11. Norm Constraints*
12. L2 (weight decay) regularization
13. Dropout
14. Label smoothing
15. BatchNorm
16. ConvNets (Pooling, Dilated convolutions**, Padding, itp)
17. Resnets (ResidualConnetions)
18. RNN
19. Vanishing and Exploding Gradients
20. LSTM
### NN Advanced
1. GoogleNet**
2. EfficientNet
3. GRU
4. Attention
5. Transformers (BERT ELMO ITP)
7. MaskRCNN
8. Object detection
9. Semantic Segmentation
10. Instance Segmentation
11. YOLO
12. GANs
13. UNETs
14. Autoencoders
15. Variational autoencoder**