Scribd
Upload
4 views1 page

Deep Learning Assignment on BNNs

This document outlines an assignment for a deep learning class involving Bayesian neural networks (BNN). It provides two problems to solve - the first asks students to derive the objective function for estimating the parameters of a BNN using posterior probability maximization for regression with Gaussian distributed outputs. The second problem asks students to do the same for binary classification with Bernoulli distributed outputs. Students taking the 4000 level class only need to attempt the first problem. This assignment will count as a programming assignment to balance homework and programming assignments.

Uploaded by

SergeiBugrov
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
4 views1 page

Deep Learning Assignment on BNNs

This document outlines an assignment for a deep learning class involving Bayesian neural networks (BNN). It provides two problems to solve - the first asks students to derive the objective function for estimating the parameters of a BNN using posterior probability maximization for regression with Gaussian distributed outputs. The second problem asks students to do the same for binary classification with Bernoulli distributed outputs. Students taking the 4000 level class only need to attempt the first problem. This assignment will count as a programming assignment to balance homework and programming assignments.

Uploaded by

SergeiBugrov
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

ECSE 4965/6965

Introduction to Deep Learning


Assignment #8
Due 4 pm, May 1

Given the discussions on Bayesian Neural Networks (BNN) in the class, let  be the parameters of a NN
and its prior distribution be specified as p()~N(0,I), where I is an identity matrix. For a NN with input X
∈RN and one dimensional output Y.

Given training data D={Xi, Yi}, i=1,2,..,M

1) For a regression problem, i.e., Y ∈R1, give the objective function that estimates  by
maximizing the posterior probability of  (not maximum likelihood estimation), assuming Y|X
follows N(f(x,), 2 (x,)), where f(x,) is the discriminant function of the last hidden layer
and 2 (x,)) is the variance of X. Show how the variance of Y is determined, given the
estimated For both questions, provide the details for each term.

2) For a binary classification problem, i.e., Y ∈ B1 and B ∈ {0, 1}, give the objective function that
estimates  by maximizing the posterior probability of  (not maximum likelihood estimation),
assuming Y|X follows Bernoulli((f(x,)), where  () is the sigmoid function. Show how the
variance of Y is determined, given the estimated For both questions, provide the details for
each term.

Note:

1) Students taking 4000 level class only need attempt problem 1.


2) To balance between homework and programming assignment, this assignment will be counted
towards a programming assignment.

You might also like