Feb 27- March 01 2013

To- Do- List:

Continue reading Anna's thesis paper.

Important Notes from the paper for me to remember:

1. Chapter 2, section 2.5, page 12.
 For the machine learning techniques, we take a feature based approach. Features are extracted from time series sensor curves[4]. Reading synchronous data from the fur sensor and the fabric pressure sensor, Anna extracted several standard sequence statistics which resulted into a memory for the creature.
The sequence features were then fed into Weka, which allowed them to compare performance of many different algorithms on their data.

2. Chapter 4, section 4.3, page 38
Gesture Evaluation and Analysis:
The key defined gestures are:
    a. stroke: moving hand gently over the prototype body, often repeatedly
    b. scratch: rubbing the prototype with fingernails
    c. tickle: touching the prototype fur with light finger movements
    d. squeeze: firmly pressing the prototype body between the fingers or both hands.
    e. pat: gently and quickly touching the body with the flat of the hand.
    f. rub: moving the hand repeatedly to and fro with firm pressure
   g. pull: gently and randomly pulling at hairs in the fur
   h. contact without movement: any undefined touch without motion
   i. no touch: prototype left untouched

3. Chapter 4, section 4.3.1, page 38:
The process of collecting data results in 4 time series curves for each example collected. Anna chose a 2 second window of observation (this was the optimal window size)
A better approach is to randomize the gesture types when collecting data!

4. Chapter 4, section 4.3.2, page 40:
Features used for classification: maximum, minimum, mean, median, area under the curve, variance, total variation. (7 items) and these features were calculated for each of the 4 time-series resulting in a total of 28 features.
The machine learning classifiers used were:
    a. Logistic regression
    b. Bayesian networks
    c. Neural networks
    d. Random Forests
used 100 fold cross validation. The data was randomly split into 100 equally-sized subsets as test data.