|
Domain |
Parts |
Remarks |
|
What is learning? |
- Intelligent behaviour requires learning or the ability to adapt & gather useful knowledge
|
In computing, it is the process of sensing & inferring new facts to update the knowledge base to be used in future
|
|
What are the forms of learning? |
- Rote learning: memorizing e.g.
- Feature extraction & induction: picking out impt. characteristics to build a predictive model
- Clustering: organize similar patterns into groups
- Online learning: agent adapting while working
- Offline learning: saving data while working & training while free
|
Supervised learning: input data + desired outputs
Unsupervised learning: input data only
Reinforcement learning: softer supervised learning where desired outputs not hard & imposed, general feedback
|
|
Neural networks? |
- Parallel, low-level cognitive network with a collection of independent neurons linked by connection weights where learning means adjustments of the weights
|
Back-propagation networks: NN using backward propagation of error algorithms
Kohonen map: unsupervised, feedforward, clustering for patterns
|
|
Others? |
- Decision trees: using information theory to split & classify data sets
|
Classifier systems: learning systems using GA to modify rule base
|