Machine Learning

Introduction

Machine learning (ML) is the study of automating a system’s use of observations to improve its ability to deal with a given task; hence, note that it is not merely the improvement of the system’s performance but its ability to perform, which implies improvement of its long-range performance. Hence, ML is not about adapting to observations in the moment but using observations to make persistent adaptations for the future. This long-range orientation differentiates ML from areas such as interactive agents (which can but do not necessarily involve ML). To elaborate, ML is the study of methods that can automatically detect patterns in data and then use the uncovered patterns to predict future data or perform other kinds of decision-making under uncertainty (which can also be planning how to collect more data) (Murphy, 2012). Given that there is inherent uncertainty in making inferences from observation without rigorous inductive reasoning, the theoretical basis of ML is the study of uncertainty, which involves the measurement and (if possible) minimisation of uncertainty; hence, probability theory is a theoretical basis for ML.

Explore

• Definitions
• Essentials of ML

Supervised learning (SL):

• Regression
• Loss Optimisation
• Supervised Learning Models

Conceptual Map

• Supervised learning (SL)
  • Regression
  • Classification
  • Key topics that pervade SL as a whole
    • Learning model
    • Objective function
    • How to choose the above
    • Optimisation using gradient descent
  • Neural network (NN) for SL
    • Theoretical basis
    • Implementation
    • Training and training hyperparameters
    • Pre-training considerations
• Bayesian inference (BI)
  … related to SL
  • Bayesian modelling
  • Components of BI
  • Approximate/computational BI
• Unsupervised learning (UL)
• Reinforcement learning (RL)

Citations

• Murphy, K. P. (2012). ‘3. Generative models for discrete data’. Machine Learning: A Probabilistic Perspective. London, England: The MIT Press. pp. 65-87.