Quiz
Module 13 Review
Create a free account to finish this quiz
Save your progress and see your final score.
Already have an account? Sign in
Questions in this quiz (14)
1Residual layers sum the information, and it results in a ______ network
- Athinner
- Bsimple
- Cfatter
- Dcomplex
2Concatenation layer combines multiple inputs by concatenating making the network _____.
- Afatter
- Bthinner
- Ccomplex
- Dsimple
3____ require fewer parameters than a fully connected layer
because the parameters are shared across columns thereby improving
computation efficiency- AConvolution layers
- BPooling layers
- CInput layers
- DFully connected layers
4_____ retains information that would otherwise be disregarded
- Apadding
- Bpooling
- CFully connected layer
- DOutput layer
5 The dimension of a feature map is determined by the input dimension, padding dimension, filter dimension and the _______.
- Astride
- BOutput dimension
- CPooling dimension
- DHorizontal dimension
6When using CNN for image classification, the _______ layer is useful when you want to check if a particular object exists in an image, more than its position.
- Apooling
- Bpadding
- Cinput
- Doutput
7 Which one of the following statements is not true about skip layers?
- AAdding a skip layer helps the network remember latent features that it may have learned early
- BSkip layers help mitigate the vanishing gradient problem.
- CA skip layer is handled through two types of layers: concatenation layers and residual layers
- DSkip layers are introduced to solve the problem of bottlenecking
8In CNN, width, height, and stride are all examples of ________.
- AHyperparameters
- BLearning parameters
- CWeights
- DFeature maps
9A______ is a stage of feature extraction that consists of multiple convolutions that are usually stacked then combined with a concatenation layer or a residual layer
- AFilter
- BBlock
- CHidden layer
- DSkip layer
10For CNN, grayscale images require ___ channel(s) while color images need ___ channel(s)
- A1, 3
- B2, 3
- C1, 4
- D2, 4
11Which of the following about filters is not true?
- AEach filter creates non-equivariant representations of the input to optimize learning
- BFilters have weights
- CFilters have a bias term
- DThe parameters in each filter are shared by multiple columns in the data.
12Which localized summary can a pooling layer provide?
- AAverage, minimum, and maximum summaries
- BOnly minimum and maximum summaries
- COnly average summary
- DNone, summaries are only done with filters, not pooling layers
13Which statements is not true about cardinality?
- AIt is the number of blocks within a transformation set.
- BIt can be expensive to train.
- CIt is an alternative to going deeper or wider for increased accuracy.
- DIf skip layers are used, residual connections are recommended to reduce training cost.
14Which of these is not a reasonable padding rule?
- AIf the stride is 2, the output image size is double the input image size.
- BPadding is added first to the right side then the left.
- CImages are padded with zeros.
- DThe horizontal and vertical padding sizes are independent of each other.
Information
- Description
- Quiz on Module 13 Review, created by Nick Feldman on 01/05/2022.
- Author
- Nick Feldman
- Language
- English
- Views
- 0
- Rating
- No ratings yet Rate this quiz