You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
38 lines
1.5 KiB
38 lines
1.5 KiB
How are things learned?
|
|
Memorization
|
|
Accumulation of facts
|
|
Limited by:
|
|
Time to observe facts
|
|
Memory to observe facts
|
|
----------
|
|
This is "declarative knowledge" - based on statements of truth
|
|
----------
|
|
Generalization
|
|
Deduce new facts from old facts
|
|
Limited by:
|
|
Accuracy of the dedeuction process
|
|
Essentially a predictive activity
|
|
Assumes that the past predicts the future.
|
|
----------
|
|
This is "imperative knowledge"
|
|
----------
|
|
|
|
|
|
Basic paradigm:
|
|
- provide a set of - seen, observed - training data
|
|
- decide on a characteristic of that training data as representative for the issue
|
|
- infer something (a rule?) about the process that has generated that data
|
|
- use inference to make predictions about previously unseen data
|
|
- confirm inference using a set of test data
|
|
|
|
A choice might have to be made between "Will I have false negatives or false positives allowed by my rules" and it would depend on what side is the risk higher.
|
|
|
|
Issues of concern when learning models:
|
|
Leaned models will depend on :
|
|
- distance metric between examples
|
|
- choice of features vectors
|
|
- constraints of complexity model
|
|
- specified or unknown number of clusters
|
|
- complexity of separating surface
|
|
- need to acoid overfitting problems like "each example is its own cluster"
|
|
|