Course structure
Introduction to machine learning
Day 1
- What is machine learning?
- Types of machine learning (classification, regression, ranking, reinforcement learning, clustering, anomaly detection, recommendation, optimization)
- Data preparation (train, test and validation data sets, imbalanced and noisy data)
- Classification model evaluation (accuracy, precision, recall, confusion matrix, ROC, AUC)
- Basic algorithms for classification (baseline models, Naïve Bayes Classifier, Logistic regression, Support Vector Machines, decision trees, ensemble models)
- Quick Scikit-Learn tutorial (how to load and transform data, training models, predicting values, model pipelines and evaluation)
- Practical classification task
- Basic algorithms for regression (analytical methods, gradient descent, SVR, regression trees)
Day 2
- Basic algorithms for clustering (K-means, hierarchical clustering)
- Practical clustering task
- Introduction to artificial neural networks (why they are so popular, what their advantages and disadvantages are, perceptron neural network)
- Most frequently used activation functions (Sigmoid, Linear, Tanh, Relu, Softmax)
- Multi-Layer neural networks (back propagation algorithm, stochastic gradient descent, convolution, pooling, regularizations)
- Quick tutorial to Keras (sequential models, optimizers, training, data workflow)
- Practical classification and regression tasks using neural networks
Convolutional Neural Networks and Image Processing
Day 3
- VGG 16 and ResNet
- Transfer learning and fine-tuning
- Image classification
- Batch normalization and data augmentation
- U-net and Image segmentation
- GANs and superresolution
- Neural network explainability
- Adversarial patch
Natural Language Processing
Day 4
- Introduction to natural language processing
- Chapters from computational linguistics (corpus, tokenization, morphological, syntactic and semantic analysis, entropy, perplexity)
- Text document vectorization (bag of words, one-hot encoding, TF-IDF)
- Practical taks on text classification
- Word embedding (word2vec, GloVe)
- Introduction to language modelling (n-gram models, smoothing, neural network based language models)
- Practical task on language modelling (implementation of a language detection algorithm based on language models)
- Neural network based text generator
Time series
Day 5
- Introduction to the theory of time series modeling
- Classical methods for time series prediction (space & frequency domain, spectral analysis, autocorrelation, ARIMA models etc.)
- Hands-on example (pandas, basic characteristics, simple prediction)
- Machine learning for time series prediction (state-space methods, Hidden Markov Chain, Kalman filter, classical neural networks, recurrent networks, LSTM)
- Hands-on examples of machine learning methods (training set preparation for specific task and model, training process & evaluation)
- Complex example of time series prediction using recurrent neural network (temperature prediction from high-dimensional input data: training data set preparation, training process & validation, prediction with trained neural network)
