Current Challenges In ML

Mature algorithms:

  • Decision trees
  • Regression
  • Neural Networks
  • Bayesian Methods
    • These can be applied to standard database relations or flat files

Some current challenges for ML are:

  • Web-Scale Data
    • The problem with large amounts of data is that when it's complexly formatted it's hard to sift through.
  • Online Learning
  • Optimisation

Potential Areas for Work:

  • Learning systems across mixed-media data (e.g. social media)
  • Learning systems across multiple databases, as well as the web and newsfeeds
  • Learning systems through active experimentation
  • Lifelong learning systems
  • Learning by analogy
    • Learning from other data to find connections between different abstract graphs