Homework 2
(due Tuesday, Feb 9th 2016)
1. Least squares implementation: Implement and execute a least squares classifier on Fishers Iris flower
dataset (http:/en.wikipedia.org/wiki/Iris_flower_data_set). The dataset contains 50 samples from
each of the th

Home Work 1: CAP 6610 Spring 15
Due Date: Feb 2nd 2015
Submit in hard copy. Show all steps. Be as concise as possible.
1. A biased coin lands heads with probability 51 each time it is flipped (i.e. the coin is biased
tails). Let X1 , ., Xn represent n con

Home Work 1: CAP 6610 Spring 15
Due Date: Feb 4th 2013
Show all steps. Be as concise as possible.
1. A biased coin lands heads with probability 51 each time it is flipped (i.e. the coin is biased
tails). Let X1 , ., Xn represent n consecutive coin flips
P

Theorem 5.1. [Bennett] Assume
Z, and t 2 D. Then
JP
13 = cfw_3,113.32 = 0'2j |Z| < M = cunts:J Z], ,3 independent copies of
(71
ZZ2t
1:].
where 415(3) 2 [1 + 3:] log(1 + .1;) 2:.
Pmo Since Z; are i.i.d.,
11 n
IP (2 Zi 2 t) S EAtEel 23:1 Z,- : EAt

Machine learning Homework 1
Let f (x) be a convex function and X cfw_x
Pi : i = 1, 2, , N is a random
variable with probabilities P (xi ) where i P (xi ) = 1. Then prove that
f (E(X) E(f (X).
Solution sketch: The proof is by induction. For the base case

Classification of Leaves
By Diganta Bharali, Cheyenne Lambert, Kara Cooper
Abstract: The project aims at studying leaf features like shape, margin and texture from
binary images and using them to identify plant species. Leaves, due to their volume,
preval

Leaf Classification Using Images
Subtitle as needed (paper subtitle)
Authors Name/s per 1st Affiliation (Author)
Authors Name/s per 2nd Affiliation (Author)
line 1 (of Affiliation): dept. name of organization
line 2-name of organization, acronyms acceptab

Lecture 5: Ensemble Learning
Course: Machine Learning for Health &
Biomedical Applications
Parisa Rashidi
Fall 2016
Ensemble Methods
Construct a set of learners from the training data
Aggregate predictions made by multiple learners
General Idea
D
Step 1

Lecture 6: Feature Extraction
and Selection
Course: Machine Learning for Health &
Biomedical Applications
Parisa Rashidi
Fall 2016
Outline
Wide data and its challenges
Feature selection
Filter-based
Wrapper
Embedded
Wide Data
Datasets with many more

Gene Expression Analysis,
Machine Learning Approach
Ashkan Ebadi
November 1, 2016
Outline
Microarray data
How to make data ready for analysis
Quality control
Normalization
Challenges
Unsupervised learning
Supervised learning
What is Gene Expressi

Exam 2 Overview
Course: Machine Learning for Health &
Biomedical Applications
Parisa Rashidi
Fall 2016
Whats left
HW4 due today
Exam on Tuesday
Final paper due Wednesday
Methods
Neural Networks
Deep Learning
Gene Expression
Ensemble
Feature Selec

Lecture 5:
Deep Learning
Course: Machine Learning for Health &
Biomedical Applications
Parisa Rashidi
Fall 2016
Deep Neural Network
Material partially based on:
-Raschka, Sebastian. Python Machine Learning (p. 18). Packt Publishing.
-Stanford CS231n: Conv

Lecture 4:
Neural Network
Course: Machine Learning for Health &
Biomedical Applications
Parisa Rashidi
Fall 2016
Methods
Decision Tree based Methods
k-NN
Support Vector Machines
Neural Networks
Neural Network
Material partially based on:
-Raschka, Seb

CAP 6610 Machine Learning: Homework 1
Due: 3rd February, Friday, 2016, In Class
1. Two pennies, one with P(head) = u and one with P(head) = w, are to
be tossed together independently. Define p0 = P (0 heads occur), p1 =
P (1 heads occur), p2 = P (2 heads

Home Work 2: CAP 6610 Spring 15
Due Date: Mar 25th 2015
Show all steps. Be as concise as possible.
1. Consider a data set with 3 points in 1-D
Class 1 : cfw_ 0
Class 2 : cfw_ 1, 1
Are the classes 1,2 linearly separable?
T
Consider mapping each point

Home Work 3: CAP 6610 Spring 15
Due Date: Apr 17th 2015
Show all steps. Be as concise as possible.
1. Derive the dual problem for slack variables version of the linear SVM. Is the form of the
dual amenable to the kernel trick? If so, write down the dual f

Homework 6
(due Tuesday, April 5th 2016)
1. LASSO implementation: Implement and execute a majorization-based regression algorithm on the
LASSO objective function written as
X
ky Xk22 +
|n |
(1)
n
This can be majorized as follows. The majorized objective

1. -margin separating hyperplanes: In linear support vector machine (SVM) learning, we are
given a set of N patterns cfw_xi , i cfw_1, . . . , N and a set of class labels cfw_yi . In the two class
problem, we have yi = 1 for i C1 and yj = 1 for j C2 . We

PROJECT DESIDERATA (UPDATED 2016/04/14)
The final project is due by Thursday April 28th at noon. A project pdf should be submitted via Canvas) with the subject title Final project: CAP6610 Spring 2016 and attachment entitled:
cap6610sp16_project_final_<in

CAP6610: Machine Learning
Midterm II
Spring 2014
UF ID #_
N
I. [50 points] Kernel Principal Component Analysis (KPCA): Given a set of patterns cfw_xn n=1
where xi RD , assume a mapping to a reproducing kernel Hilbert space (RKHS) giving us a set of patter

CAP6610 Spring 2016
HW5 Solutions
12.12 Show that the Kullback-Leibler divergence KL (pkq) is a nonnegative quantity. Hint.
Recall that ln () is a concave function and use Jensens inequality, that is,
g (x) p (x) dx f (g (x) p (x) dx
f
where p (x) is a

CAP6610 Spring 2016
Midterm 1 Solutions
ML 1. [40 points] Support Vector Machine:
In a version of the one versus all multiple class linear support vector machine, the following objective
function is minimized.
K
EMCSVM (cfw_k , k0 ) =
K
h
i
XXX
1X
T
kk k2

CAP6610 Spring 2016
HW1 Solutions
6.1 Show that if A Cmm is positive semidefinite, its trace is nonnegative.
By the definition of a nonnegative matrix, for x Cm , we have xH Ax 0. Hence, this will also
be true for x = [1, 0, 0, ., 0] so that xH Ax = [A]11

CAP6610 Spring 2016
HW3 Solutions
11.5
Since the kernel matrix K in order N is positive semidefinite, when let N = 2, we have
(
K=
)
(x, x) (y, x)
(x, y) (y, y)
and it satisfies T K 0 for any . Assuming = [ (y, y) , (x, x)]T , there is
[
]
(x, x) (y, y

SHARP QUADRATIC MAJORIZATION IN ONE DIMENSION
JAN DE LEEUW
A. Quadratic majorizations for real-valued functions of a real
variable are analyzed, and the concept of sharp majorization is introduced.
1. I
Majorization algorithms, including the EM algorithm,

CAP6610: Machine Learning
Midterm I
Spring 2014
Student UF ID #_
ML 1. [40 points] -margin separating hyperplanes: In linear support vector machine
(SVM) learning, we are given a set of N patterns cfw_xn , n cfw_1, . . . , N and a set of class labels
cfw

Homework 4
(due Tuesday, March 8th 2016)
1. SVM implementation: Implement and execute a majorization-based two class linear SVM on Fishers
Iris flower dataset (http:/en.wikipedia.org/wiki/Iris_flower_data_set). The dataset contains 50
samples from each of

An Empirical Comparison of Supervised Learning Algorithms
Rich Caruana
Alexandru Niculescu-Mizil
Department of Computer Science, Cornell University, Ithaca, NY 14853 USA
Abstract
A number of supervised learning methods
have been introduced in the last dec

Home Work 2: CAP 6610 Spring 2017
[Programming Assignment]
Due Date: March 15th,Wed
Please submit submit a zip file which has all the dataset, report and codes for
the homework.We will not accept any submission without a zip file.
1. Single Node Neural Ne