This chapter starts with a brief historical introduction to neural networks and the basic
perceptron, basically a neural network without hidden layers. We also discuss the
multilayer perceptron and training with gradient decent.
The historical t
In the previous learning problems we had training examples with feature vectors x
and labels y. In this chapter we discuss unsupervised learning problems in which no
labels are given. Training on unlabeled examples restricts the ty
Regression and maximum likelihood
We have so far given an overview of supervised learning where we propose a model
in form of a parameterized function and where we then learned the model parameters from example data. We have discussed this with a simple
This chapter follows closely: http:/cs229.stanford.edu/notes/cs229-notes2.pdf
MLE of Bernoulli model
An important special case of learning problems is classification in which features are
mapped to a finite number of possible categori
Due Oct 17, 2016, 4pm by email to email@example.com with subject line A4. The
page limit for this assignment is 2.
What is the goal of supervised machine learning?
Explain briefly what k-fold cross validati
Due Nov 16, 2016, 4pm by email to firstname.lastname@example.org with subject line A6.
CSCI4155 (undergraduate students only):
1. Write a program to demonstrate k-mean clustering as in the example
shown in class. The example points should b
Due Oct 10, 2016 by email to email@example.com with subject line A3
The file train.txt on our course wiki contains data with 11 rows and 200
columns representing 200 examples of data with 10 feature values and 1
CSCI 4155/6508 Machine Learning 2016: Assignment 2
Please note that late assignments are not accepted. Also, I recommend that you start
with the assignments early as I expect that you have questions. Past experience has
shown that starting assignments