Graphical models

Neural network: re-engineering algorithms. The human brain is the most amazing machine learning algorithm ever invented.

Support vector mechanism

Evaluation

• Accuracy
• Precision and recall
• Squared error
• Likelihood
• Posterior probability
• Cost / Utility
• Margin
• Entropy
• K-L divergence

Types of Learning

• Supervised (inductive) learning
• Training data includes desired outputs
• Unsupervised learning
• Training data does not include desired outputs
• Semi-supervised learning
• Training data includes a few desired outputs
• Reinforcement learning
• Rewards from sequence of actions

Inductive Learning

Given examples of fucntion (X, F(X)), where X is the input und F(X) the output. If the outcomeis discrtete. It is

• Discrete F(X): Classification
• Continues F(X): Regression
• F(X) = Probability(X): Probability estimation

Probality is a special type of regression. probability needs to sum up to 1.

Learning Methods

• Supervised learning

  • Decision tress induction
  • Rule induction
  • Instance-based learning
  • Bayesian learning
  • Neural networks
  • Support vector machines
  • Model ensembles
  • Learning theory

• Unsupervised learning

  • Clustering
  • Dimensionality reduction

Machine Learning in Practice

• understand domain, prior knowledge, and goals
• Data integration, selection, cleaning, pre-processing, etc.
• Learning models
• Interpreting results
• Consolidating and deploying discovered knowledge
• Loop

The first thing to do is to understand the domain, learn the biology. Learn natural language if you want to process documents. It is also important to understand what the goals are. Sometimes the goals are very clear. We have more things to try than we can. So in sum it is very important to prior understand the domain, the goal and constraints.

The next step is the most time consuming. We need to get the actual data. No data, no learning. Big data, big learning and huge opportunities. The more data you have the more crap you have. You have to find your data, you have to clean it and integrate it. You often have multiple sources of data. In all large projects, this is always the case. There is rarely a single source of data, but rather multiple sources of data often with very different origins and you have to put them all together. If you put them all together wrong, then its like garbage in and garbage out.

Machine learning is a cycle that has many steps. 6:35

Inductive Learning

We are given a d training example of some unknown function x. The training example is a pair of two things, an input (e.g. text) Simple binary decisions.

Appropriate Applications for Supervised Learning

• Situation where there is no human expert

• x: Bond graph for a new molecule

• f(x): Predict binding strength to Gerontogene

• Situations where humans can perform the task but can not describe how they do it

• x: Bitmap picture of hand-written character

• f(x): Ascii code of the character

• Situations where the desired function is changing frequently

• x: Description of stock prices and traces for the last 10 days

• f(x): Recommended stock transactions

• Situation where each user needs a customized function f

• x: Incoming email message

• f(x): Importance score for presenting user (or deleting without presenting.

Is changing every day

The essence of Inductive Learning

If ther eis no answere to a question

Makes discovery 26:00

Tags: ai