cs571_Week_4_Lecture_questions_1

pdf

School

University of Nevada, Las Vegas *

*We aren’t endorsed by this school

Course

571

Subject

Computer Science

Date

Jun 11, 2024

Type

pdf

Pages

Uploaded by SuperInternet12237

How does dropout approximate multiple neural network configurations from a single network? By stochastically zeroing the neurons of a layer. By stochastically zeroing the layers of a network. By zeroing each subset of neurons i a sequence. By zeroing each neuronin a sequence. © comect Correct! This effectively results in multiple differently trained networks, which at inference time are averaged together. 1fa neural network can be considered as a function mappinginputs to outputs, what can a network using Monte Carlo Dropout be considered as? Aprobability density function Asingle function Astepwise function Asetoffunctions © comect Correct! Specifically, a set of functions each with an independent mapping of input to output. When making stochastic forward passes through network using Monte Carlo Dropout, when do we update which neurons are randomly dropped out? Once for every new input sample Once after training, before inference Randomly throughout the forward pass Before every stochastic forward pass © comect Correct! This effectively gives a new network configuration from which each forward pass to build a distribution. 1. Which type of dropout drops neurons during both training and inference? 171 point O Posterior Dropout O Universal Dropout @ Monte Carlo Dropout O sanDiego Dropout ! This technique, which approximates the mean of many different neural networks, was introduced by Yarin Gal and Zoubin Ghahramani in 2016.

2. What makes neural networks in general different from networks with Monte Carlo Dropout? 1/1point @ Inference s deterministic O The number of neurons i the network architecture is constant. (O The number o lsyers in the network srchitecture s constant. O inference s nondeterminisic @ comrect Correct! With networks using Monte Cario Dropout, neurons are stochastically dropped out at inference, resulting in nondeterministic outputs. 3. Whatis the output of a neural network that uses Monte Carlo Dropout? O Adistrbution sround the trgetvslus. @ Asetofinferences O Asingle more sccurstz inference. O Aninference as well s s cerainty. © comeet Correct! We can then, for example, treat this set as a normal distribution and use the mean of this distribution as our prediction and the variance as a metric of its certainty in that prediction. 4. When using Monte Carlo Dropout, which element of the distribution of outputs provides information about the 171 point modelscertsinty? O The dropout prabsbity O Themesn @ Thevarisnce O Themedian © comeet Correct! When viewing the outputs as a norma distribution, a lower variance implies a higher certainty. 5. Suppose that a neural network used some kind of dropout. What should you check for in order to determine 171 point whether regular Dropout or Monte Carlo Dropout was used? @ rdrpout s used stinference. O frdrpout s used sttsining. O frdropout s spplied st every hidden lyer. O Ifaconstantdropout probabilty s used across layers. © comeet Correct! Thisis the key difference that separates Monte Carlo Dropout from regular Dropout, and it allows s to make nondeterministicinference to generate a probability distribution of outputs.

Why s an approach like convolutional neural networks better suited for object recognition in digital images than the traditional programming paradigm? There s no straightforward mapping of input images to classifications. There s no other way to extract necessary information such as outlines. Digital images are too high resolution. Traditional methods would be too slow: © comect Correct! There is not a clear method to take in the numeraus pixel values and compute and return a classification from them. What form does data for an RGB digital image take? 3vectors Asingle matrix Alist of 3-tuples A 3-channel matrix ) correct Correct! RGB digital images have a separate channel for each color: red, green, and blue. Which component of a CNN involves procedurally running kernels over regions of pixels in a digital image? Fully connected layers Convolutional layer Pooling layer Flattening Correct! The convolutional layer will output a feature map for each kernel applied. How do the values of an image change after applying Gaussian Blur? Smaller values throughout Smaller values in the middle and higher at edges Higher values in the middle and smaller at edges Higher values throughout @) correct Correct! The Gaussian filter attributes larger weight to nearby pixels than remote pixels.

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

In image processing, how were kernels originally created? There is known set of possible kernels. Kernels were originally created by CNNs. Kernels were originally created by ANNs. Kernels were originally created by human design. (@) Correct Correct! With a convolutional neural network, we can use machine learning to find useful kernels instead. How does a kernel transform an image? By reducing the pixel width and height By forward propagating pixel values through fully connected layers By linearly combining pixel regions By scaling pixel regions () correct Correct! If we look at the kernel and pixels as a vector, the new pixel value becomes their dot product. What does a CNN learn in a convolutional layer? Auseful kernel for each feature map How to connect feature maps into a classification Auseful kernel for each pixel region Weights connected to every pixel @ comeet Correct! Convolutionalneural nebworlks are mch more eficent b leaning kemels that can be universally applid tothe enireImage instead of weights for evry Indvidual pixel and channeL. What does “stride” refer to in a convolutional layer? Zero-valued pixels added to the edges of the image Horizontal pixel step size of each convolution Horizontal and vertical pixel step size of each convolution Size of pixel region for each convolution ) correct Correct! Stride determines how many pixels are shifted between each convolution.

Within a convolutional neural network, how are different feature maps generated for the same image? Convolving a kemel to different regions of the image. Max pooling the previous feature maps Convolving different kernels. Varying weights in the fully connected layers. © correct Correct! A feature map will be created for each kernel applied. Ina CNN, what constitutes 3 typical repeatable convolutional component? Flattening > Fully connected layers Pooling > Flatiening Convolutionsl layer-> Flstiening Convolutionsllayer > Pooling © comect Correct! These two steps form common unit, which can b repeated when designing a network's architecture. Which task must be performed on the feature maps in order to pass them to fully connected layers? Flattening Convolving Zero-padding Striding @ correct Correct! Each value within the feature maps s stacked into a vector to be passed through a fully connected network.

When is it *most” necessary to use a pretrained convolutional neural network? When there is a larger network architecture When there are only nonstandard classes When there is limited data with which to train When higher accuracy is needed (¥) Correct </ Correct! Convolutional neural networks require great volumes of information to be effectively trained. With limited data, it is possible to use a pretrained network and then train further on a limited dataset. How does visualizing the activation maps in a CNN help? Decide when convergence has reached Calculate the loss Hints at what the CNN is focusing upon Define what the CNN is focusing upon ‘G/) Correct Correct! Visualizing activation maps in Convolutional neural networks does not give us any definite ideas but just hints as to what might be going on inside the CNN. N

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Related Documents

SPN 100 - Final Exam Study Guide.pdf

Assignment3 (1).docx

Assignment4 (1).docx

Crafting Effective Lesson Plans for Workshop Success

cs571_Week_3_Lecture_questions_2.pdf

CoreTech2-3MilestoneOne.docx

cs571_Week_2_Lecture_questions_1.pdf

Practice 08-42 Comparac....pdf

Module 3 Assignment Martinez.docx

GTQ Lab Report 1.docx

Module 6 Assignment Martinez.docx

Module 9 Assignment Martinez.docx

Recommended textbooks for you

Operations Research : Applications and Algorithms

Computer Science

ISBN:9780534380588

Author:Wayne L. Winston

Publisher:Brooks Cole

Principles of Information Systems (MindTap Course...

Computer Science

ISBN:9781305971776

Author:Ralph Stair, George Reynolds

Publisher:Cengage Learning

MIS

Computer Science

ISBN:9781337681919

Author:BIDGOLI

Publisher:Cengage

Principles of Information Systems (MindTap Course...

Computer Science

ISBN:9781285867168

Author:Ralph Stair, George Reynolds

Publisher:Cengage Learning

Fundamentals of Information Systems

Computer Science

ISBN:9781305082168

Author:Ralph Stair, George Reynolds

Publisher:Cengage Learning

SEE MORE TEXTBOOKS

Recommended textbooks for you

Operations Research : Applications and Algorithms
Computer Science
ISBN:9780534380588
Author:Wayne L. Winston
Publisher:Brooks Cole
Principles of Information Systems (MindTap Course...
Computer Science
ISBN:9781305971776
Author:Ralph Stair, George Reynolds
Publisher:Cengage Learning
MIS
Computer Science
ISBN:9781337681919
Author:BIDGOLI
Publisher:Cengage
Principles of Information Systems (MindTap Course...
Computer Science
ISBN:9781285867168
Author:Ralph Stair, George Reynolds
Publisher:Cengage Learning
Fundamentals of Information Systems
Computer Science
ISBN:9781305082168
Author:Ralph Stair, George Reynolds
Publisher:Cengage Learning