DISCnet Machine Learning Course

Notes, demos and materials for learning Machine Learning

Extra materials

In addition to the material in this git repository, I've also used materials from my computer vision, data mining and deep learning modules. Please feel free to take a look at the lecture slides and notes for these which can be found here:

• http://comp3204.ecs.soton.ac.uk / https://github.com/jonhare/COMP3204
• http://comp6237.ecs.soton.ac.uk / https://github.com/jonhare/COMP6237
• http://comp6248.ecs.soton.ac.uk / https://github.com/jonhare/COMP6248

Rough Plan

(Note that this is only a guide. We'll adapt the content to your needs during the course.)

• Monday: Overview of Machine Learning
  • Leaders: Prof Niranjan, Prof Prugel-Bennett and Dr Hare
  • Topics Covered:
    • Success stories in machine learning
    • Failures of machine learning
    • Machine learning techniques
      • Linear Regression, MLP, SVMs, Decision Trees, Deep Learning
    • Machine learning problems
      • Supervised learning (regression/classification), Unsupervised learning (PCA/clustering), Semi-supervised learning, reinforcement learning
    • Making sense of data
      • Types of data (images, text, numbers)
        • Encoding data and feature extraction
      • Data preparation, missing data
        • Balancing data
    • Common tools
    • Matlab, python
    • Practical walkthrough - understanding variance
    • Machine Learning 101 - classifying text
    • Discussion about what problems we would like to explore on Friday
    • Homework - reflect on how the day's activities could be applied
• Tuesday: Introduction to Machine Learning
  • Leaders: Prof Niranjan
  • Topics Covered:
    • The perceptron/Bayes optimal decisions
    • Feature selection and Lasso
    • MLPs
    • Gradient learning, SGD, momentum
    • Evaluating performance
      • ROC curves
    • Homework
• Wednesday: Advanced Machine Learning
  • Leader: Prof Adam Prugel-Bennett
  • Topics Covered:
    • Generalisation
      • Bias-Variance Dilema
    • Ensemble Techniques
      • Ada-boost, random forest
    • Kernel methods
      • SVM
      • kernels
    • Probabilistic techniques
      • Gaussian Processes
      • Graphical Models, LDA, MCMC
    • Homework
• Thursday: Deep Learning
  • Leader: Dr Jonathon Hare
  • Topics Covered:
    • Why Deep
      • CNNs
      • RNNs (LSTM, etc.)
    • Word Embeddings
    • Loss functions
    • GPU programming (libraries)
    • Keras tutorial 1 - building simple CNNs
    • Transfer Learning
    • Keras tutorial 2 - transfer learning with CNNs
    • Keras tutorial 3 - Text classification
    • Keras tutorial 4 - Sequence modelling
    • Current research challenges
      • Visual
        • segmentation
        • object detection
        • multi-label classification
      • Text
        • sequence-sequence learning
          • translation, embedding, etc
          • logical inference & QA
      • Cross-modal transfer
        • generating from embeddings
        • VQA
      • GANs
    • Homework
• Friday: Practical Machine Learning
  • Leaders: Prof Niranjan, Prof Prugel-Bennett and Dr Hare
    • Workshop on data you provide
    • We will look at:
      • Analyse the problem
      • Visualise the data
      • Cleaning the data
      • Using machine learning libraries
      • Evaluate performance