ME 537: Learning Based Control
Week 1, Lecture 1: Introduction
HW 1 Due on 10/8
Oregon State University!
! Introduction ! Syllabus ! About course
! Topics ! Expectations ! Project
! Questions ?
Oregon State Uni
A 100(1 )% simultaneous condence interval for i j is given by
(i yj ) HSDij .
Example. GPA/residency data. Here, r1 = r2 = r3 = 30. So:
LSD = t /2, N t
= t 0.025, 87
= 1.988 0.1806 = 0.359
q 0.05 3, 87
q , t
Analysis of covariance
Note: Kuehls discussion of ANCOVA, on pp. 550-564, is somewhat dierent
from mine. Focus on the concepts, and on the specic issues I cover in lecture.
ANCOVA combines a qualitative independent variable (factor) with a qu
Analysis (one possible approach)
1. Start with the most complicated model (Model 3), and test H0 : 4 = 5 = 0
(equality of slopes ). Under H0 , Model 3 reduces to Model 2.
SSE r SSE
38.57 31.52 31.52
= 1.01 (P = 0.4)
Random eects models
So far weve dealt with xed eects models: the treatments are specically chosen
by the experimenter, and we wish to test hypotheses about the treatment means.
e.g., three drugs inuencing blood pressure; cotton content inuenc
Adjust the mean to the value expected at x (the mean of the x values):
yi (adj) = yi (i x ),
where is the common slope estimate. See the illustration on p. 35.
For the cracker data (using x = 25, and 3 = 0.90, the common slope estimate
Randomization is restricted, so that each treatment occurs once in each block.
For each block, obtain a random permutation of the numbers 16, e.g., using
sample(1:6) in R. Assign treatments to plots within blocks, usin
ME 515: RISK AND RELIABILITY BASED DESIGN
Fall 2010 Term Mondays and Wednesdays 2:00-3:50pm ROG 332 Course Web Site: http:/classes.engr.oregonstate.edu/mime/fall2010/me515/ Instructor: Professor Irem Y. Tumer Rogers 408 Irem.firstname.lastname@example.org 541-737-
ME 537: Learning-Based Control
Week 1, Lecture 2 Neural Network Basics
HW 1 Due on 10/8 Data sets for HW 1 are online Project selection 10/11
Suggested reading :
NN survey paper (Zhang) Chap 1, 2 and Sections 4.1 to 4.5 in Passino
Random eects model
Example. Measurements of lipids in replicate serum samples run on four machines
on each of four days (Kuehl, p. 233). Because the four machines can be thought
of as a random sample from a larger population of machines, and