Machine Learning NG 2014 Coursera

<p>Sep 24, 2014 10:10 PM

I sign up (to watch lectures only)</p>

<p>

lecture 1 Welcome 7 mins: &nbsp;ML;

want to build AI at human level but we are a long ways away.</p>

<p>

ex: &nbsp;DB mining; &nbsp;web click data; med records, biology, engr.</p>

<p>

autonomous helicopter &nbsp;had to learn by itself;</p>

<p>

handwriting recog; &nbsp;most of NLP and computer vision.</p>

<p>

self-customizing programs: &nbsp;Amazon; &nbsp;Netflix recoms;

Understand the brain;</p>

<p>

* &nbsp;intro... what is ML *</p>

<p>

Arthur Samuel &nbsp;1959 &nbsp;... &nbsp;learn w/o being explicitly programmed.

checkers... &nbsp;&nbsp;could beat Art Samuel.</p>

<p>

Tom Mitchell &nbsp;&nbsp;program is said to learn from exper E with respect to T

if &nbsp;performance P improves with experience E.</p>

<p>

supervised learning: &nbsp;teach it

unsupervised... it learns by itself</p>

<p>

also RL &nbsp;and recommender systems.</p>

<p>

&nbsp;supervised learning </p>

<p>

portland oregon....</p>

<p>

Y axis &nbsp;price in K bucks

X axis &nbsp;&nbsp;size in feet2.</p>

<p>

fit a line or a quadratic curve.... &nbsp;&nbsp;supervised.

right answers are given...</p>

<p>

perhaps use regression:</p>

<p>

breast cancer</p>

<p>

Y axis: &nbsp;Yes malignant (1) &nbsp;versus No benign = 0; &nbsp;&nbsp;versus

X axis &nbsp;tumor size</p>

<p>

what is probability that say 3.5 cm is benign...</p>

<p>

classification: &nbsp;&nbsp;0 or 1....</p>

<p>

or classify &nbsp;0,1, 2, or 3.</p>

<p>

put on one axis (just map to X axis with O and X.</p>

<p>

&nbsp;

perhaps age on Y axis and tumor size on X axis.</p>

<p>

straight line separates the two populations</p>

<p>

two features....</p>

<p>

clump thickness; &nbsp;uniformity of cell size; uniformity of cell shape;</p>

<p>

or an infinite # of features...</p>

<p>

SVM &nbsp;&nbsp;there is a trick to do it........</p>

<p>

&nbsp;unsupervised learning *</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>

<p>

&nbsp;</p>