EE798L: Topics

• Linear Modeling: A Least Squares Approach

  • Linear modeling

  • Generalization and over fitting

  • Regularized least squares

  • Wireless application - 5G massive MIMO zero-forcing receiver design

• Linear Modeling: A Maximum Likelihood Approach

  • Errors as noise – thinking generatively

  • Maximum likelihood

  • Bias-variance trade-off

  • Effect of noise on parameter estimates

  • Wireless application - 5G massive MIMO MMSE receiver design

• The Bayesian Approach to Machine Learning

  • Exact posterior

  • Marginal likelihood

  • Hyperparameters

• Bayesian Inference

  • Non-conjugate models

  • Point estimate – MAP solution

  • Laplace approximation

  • Wireless application - 5G massive MIMO channel estimation

• Classification

  • Probabilistic classifiers – Bayes classifier, logistic regression

  • Non Probabilistic classifiers- K-nearest neighbors

  • Discriminative and generative classifiers

  • Wireless application - Detection in digital communication systems

• * Sparse kernel machines

  • Support vector machines (SVM)

  • Sparse Bayesian learning (SBL)

  • Wireless application

    • SVM for beamforming and data detection in millimeter wave systems

    • SBL for channel estimation in 5G massive MIMO wireless systems

• Clustering

  • General Problem

  • K-means clustering

  • Gaussian mixture models (GMM)

    • EM algorithm – MAP estimates, Bayesian mixture models

  • Wireless application - Clustering for 5G massive MIMO wireless systems using K means and GMM

• Principal components analysis and latent variable models

  • General Problem

  • Latent variable model

  • Variational Bayes

  • Probabilistic model for PCA

  • Wireless application - Variational Bayes for 5G massive IoT device detection

• Gaussian Processes

  • Gaussian Processes for regression

  • Gaussian Processes for classification

  • Wireless application - Power allocation for 5G network