Course Syllabus & Schedule
Logistics, detailed topics, and session plan.
Logistics
- Timings: 2.00 - 3.30 pm (Tuesdays and Thursdays)
- Place: CDS 102
Topics
Foundations of machine learning
- Review of ML fundamentals: Un/semi/self-supervised learning, feature-based clustering, model fitting, linear regression, Discriminative models: logistic regression, discriminant analysis basics (2 weeks)
- Unsupervised/supervised ML techniques: Clustering techniques; Expectation maximization (EM) – K-Nearest Neighborhood classifiers, Gaussian Mixture Models; Representation learning; Supervised ML methods: kernel-based methods – support vector machines, ensemble methods: Classification and regression trees (CART), boosting/bootstrap aggregation. (4.5 weeks)
- Dimensionality reduction techniques: Principal component analysis (PCA), linear discriminant analysis (LDA). (1.5 week)
- Deep learning basics: Computational graphs, feedforward networks, loss functions, convolutional neural networks, backpropagation, optimization, feature saliency and visualization, convolutional neural networks, encoder-decoder models, graph-based models, generative models. (2 weeks)
ML applications and statistical evaluation
- Classification, segmentation and decision-making: Template matching, correlation – audio/speech signals; Regression and classification on publicly available sensory/survey data, image segmentation & classification – machine learning classifiers, feature-based and rule-based decision making, uncertainty estimation. (1 week)
- Evaluation of analysis tasks: Evaluation metrics, segmentation evaluation metrics (IoU, Dice, Jaccard indices, Hausdorff distance measures), classification evaluation metrics (confusion matrix, sensitivity, specificity, accuracy), registration metrics (MSE, MAE). (1 week)
- Testing statistical significance of ML applications: Review of hypothesis testing basic, permutation tests, effect of sample size, Descriptive statistics (mean, standard deviation, median, confidence interval, IQR), Unpaired and paired t-tests. (2 weeks)
Practicals
- Implementation: Linear and logistic regressions, least square fitting, under/overfitting, regularization, LDA.
- Scikit-learn: ML classifiers, comparison of unsupervised/supervised ML classifiers on data, feature ranking, feature reduction.
- PyTorch/TensorFlow: Deep learning parameter tuning, training regime, feature visualisation/saliency.
- Python: Statistical tests (t-test).