Machine Learning 2: predictive models, deep learning, neural network 2400-DS2ML2

1. Decision trees in regression: methods of building trees, splitting algorithms CART, C5.0, splitting criterion, interpretation of results, problem of overfitting, stopping criteria, prunning the trees, cost-complexity parameter, cross-validation, testing, building predictions using trees,
2. Decision trees in classification: splitting criterions and measures of node impurity, Gini coefficient, Information Gain, node entropy, primary and surrogate splits,
3. Bagging and random forests: boostrapping and resampling methods, sampling with replications, subagging, predictor randomization, averaging the trees. out-of-bag errors, parameters tuning
4. Boosting: weak learners vs. Strong learners, gradient boosting, extreme gradient boosting, regularization methods, adaptive boosting
5. Ensembling: simple and weighted averaging, majority voting, weighted voting, stacking of bottom-layer models and top-layer model
6. Neural Networks: artificial neurons, topology of the neural networks, input layers, hidden layers, output layers, weights, bias, activation functions, backpropagation algorithm, methods of weights modification
7. Convolutional Neural Networks: filters, receptive fields, activation maps, structure of CNN, convolutional layers, ReLu layers, pooling layers, dropout layers, stride, padding, transfer learning, methods of data augmentation, backpropagation algorithm
8. Recurrent Neural Networks: sequence processing, idea of internal loop, recurrent connection, state variable, vanishing gradient problem, LSTM/GRU layers, carry dataflow, recurrent dropout, recurrent layers stacking, bidirectional recurrent networks, time-series forecasting using RNNs, sequences classification, sentiment analysis