Assignment #$1$ : Building a Convolutional Neural Network $-$ Model $2$ Architecture

Submitted on :$26$th September $2024$ @ $11$:$59$ PM Total mark $(20\%)$

Objective:

The goal of this assignment is to build and compile a convolutional neural network $($CNN$)$ with four convolutional block, each having a Conv$2$D$,$ MaxPooling$2$D$,$ and a Dropout layer. You will also initialize an Adam optimizer and compile the model using the following configurations:

Assignment Instructions:

$1.$ ModelArchitecture:ImplementtheCNNarchitecturewiththefollowing

configuration:

o Conv$2$D Block $1$:

$$ Conv$2$Dlayerwith$256$filters,a$5$x$5$kernel,'same'padding,and'relu' activation.

$$ Inputshape$=(150,150,3)$forthefirstConv$2$Dlayer.

$$ MaxPooling$2$Dlayerwitha$2$x$2$kernelandstridesizeof$2$x$2.$

$$ Dropoutlayerwitha$0\mathrm{.}25$dropoutratio.

o Conv$2$D Block $2$:

$$ Conv$2$Dlayerwith$128$filters,a$5$x$5$kernel,'same'padding,and'relu'

activation.

$$ MaxPooling$2$Dlayerwitha$2$x$2$kernelandstridesizeof$2$x$2.$ Dropoutlayerwitha$0\mathrm{.}25$dropoutratio.

o Conv$2$D Block $3$:

$$ Conv$2$Dlayerwith$64$filters,a$3$x$3$kernel,'same'padding,and'relu'

activation.

$$ MaxPooling$2$Dlayerwitha$2$x$2$kernelandstridesizeof$2$x$2.$ Dropoutlayerwitha$0\mathrm{.}25$dropoutratio.

o Conv$2$D Block $4$:

$$ Conv$2$Dlayerwith$32$filters,a$3$x$3$kernel,'same'padding,and'relu'

activation.

$$ MaxPooling$2$Dlayerwitha$2$x$2$kernelandstridesizeof$2$x$2.$ Dropoutlayerwitha$0\mathrm{.}25$dropoutratio.

$2.$ FlattenandFullyConnectedLayers:

o Add a Flatten layer to convert the output of the final convolutional block into

a $1$D tensor.

o Add a fully connected Dense layer with $64$ neurons and 'relu' activation. o Add another fully connected Dense layer with $32$ neurons and 'relu'

activation.

o Add a final Dense layer with $3$ neurons and 'softmax' activation $($for multi$-$ class classification$).$

$3.$ OptimizerandModelCompilation:

o Initialize an Adam optimizer with a learning rate of $0\mathrm{.}001.$ o Compile the model using:

$$ Lossfunction:categorical$\_$crossentropy.

$$ Metrics:accuracy.

$4.$ ModelSummaryandObservations:

o After building and compiling the model, print the model summary.

o Write down your observations based on the model architecture, including

the number of parameters in each layer and the overall complexity of the

model. $5.$ Submission:

o Submit your Python code that implements the above architecture. o Provide your observations along with the model summary.

Assessment Criteria:

$1.$ Correctness$(50\%)$:

o The model architecture is correctly implemented as per the instructions. o All layers and configurations $($filters$,$ kernel size, dropout ratio, etc.$)$ are

appropriately defined. $2.$ ModelCompilation$(20\%)$:

o The model is compiled correctly with the specified optimizer, loss function, and metric.

$3.$ Observations$(20\%)$:

o Clear and insightful observations are provided based on the model

summary.

$4.$ CodeOrganizationandClarity$(10\%)$:

o The code is well$-$structured, with proper comments and explanations for each step.

Optional Extension:

For additional credit, you can experiment with different learning rates, optimizers, or activation functions and observe how these changes affect the model's performance.

General Tips for Submitting Python Assignments:

$$ Comment your code: Include comments explaining your logic and code structure to make it easier for your instructor to understand.

$$ Test your code: Ensure the Python script runs without errors before submitting.

$$ Submit on the blackboard $2$ files format in the original code with compiled

results in: $.$ipynb and PDF file