# Week5 - What went right Week 5 Quiz 1 Redux Learning Vision...

This preview shows pages 1–4. Sign up to view the full content.

Week 5 Quiz 1 Redux Learning Vision Chapter 20, 24 Quiz 1 ! In general, grades a bit lower than I expected ! Average: 68 (!CSEgrad: 57) ! Min: 35 ! Max: 100 (!CSEgrad: 88) ! Spread is quite high ! Problems 1,5 were the worst offenders Stats Everybody Non-(CSE-grads) What went right ! Almost everyone got ! Inference cloud problem (Problem 4) ! Many people got ! Figuring out how to break down probabilities into components ! Probability distributions What went wrong ! Many people had trouble with ! Double Markov Model problem (prob 1) ! Values versus variables (prob 5) Question-by-question breakdown

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
Back to machine learning Evaluating Learning ! Training data must be selected so as to reflect the global data pool ! Testing on unseen data is crucial to prevent over-fitting to the training data : ! Unintended correlations between input and output Eg. Photos with tanks were taken on a sunny day ! Correlations specific to the set of training data Eg. Language processing trained on Wall Street Journal articles may not extend to spoken dialog Training vs. Test data ! Learning agents are presented a collection of training examples ! Modifications are made to the learning algorithm until it performs well on the training data ! Test data is held out from this process ! When performance on training data is acceptable, algorithm is run on test data ! Only performance on test data is reported Test data methodologies ! Single pass: reserve x% of data for test ! Cross-validation: ! Each fold reserves x% of data, trains on rest and tests on held out data ! Performance is averaged across folds ! Statistical tests tell how many test cases are needed for reliable conclusions Fold 1: Fold 2: Integrating CV with Decision Trees ! As a tree gets deeper, hypothesis becomes more specific ! If too specific, then hypothesis overfits training data ! For each CV fold, train full d-tree on training set ! Then prune back tree to minimize error on CV set ! Find best depth d across all folds ! Regrow tree using full data set, but stop at depth d. Ensemble learning ! Develop a suite of classifiers and combine their votes (pick the majority classification) ! Train a new classifier on the errors from another classifier
Boosting ! Training examples are assigned weight or importance ! The classifier’s opinion is weighted in proportion to the weight of the examples it learned ! Initially all examples weighted the same ! Weight of misclassified ones are boosted Boosting Non-parametric learning ! Neural networks (and Gaussians as we will see later) are parametric learners ! Restricted number of parameters according to a particular form weights, means, variances ! Non-parametric learners use the data directly to derive classifications ! Nearest neighbors / k-nearest neighbors Nearest neighbor k-nearest neighbor ! Idea: points are likely to be clustered together ! Classification: similar classes cluster together ! Probability density: instances likely to cluster together ! To figure out what to do with a point, look at

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 04/13/2010 for the course CSE 730 taught by Professor Ericfosler-lussier during the Fall '08 term at Ohio State.

### Page1 / 12

Week5 - What went right Week 5 Quiz 1 Redux Learning Vision...

This preview shows document pages 1 - 4. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online