CS7643 Quiz 2 Exam (2025) – Certified Questions
with 100% Verified Answers
INTRODUCTION:
This document contains the certified exam questions and verified
answers for CS7643 Quiz 2 (2025). It covers deep learning
architectures, CNNs, optimization, error analysis, transfer learning,
adversarial images, convolutional operations, segmentation models
(LeNet, AlexNet, VGG, ResNet, U-Net), and object detection methods
(YOLO, RPN, Mask R-CNN). The file also includes an extended study
guide with 60+ extra Q&As on neural networks, optimization
techniques, regularization, RNNs, LSTMs, GANs, reinforcement
learning, and SVMs.
EXAMS QUESTIONS AND ANSWERS:
Evolving architecture --- correct answer ---Architecture is the
subject to learn
>> evolutionary or reinforcement learning architectures
>> pruning non efficient connections
>> non-easy to optimize
,EfficientNet (ENet) --- correct answer ---CNN architecture that
operate with less parameters but perform very well.
Optimization error --- correct answer ---The model can not find the
best weights.
Higher for more complex NN with many parameters.
>> residual NN
>> evolving architectures
Estimation error --- correct answer ---Not proper generalization or
the model overfits.
Higher for more complex models.
>> regularization
>> dropout layer
Modeling error --- correct answer ---Not learning the pattern in the
dataset.
Higher for very simple NN.
--> increase the capacity
--> create more complex model
Transfer learning --- correct answer ---Reuse the features on a new
dataset, that were learned previously on a large-scale dataset.
,1. Train features on a large-scale dataset
2. Replace the last FC layer with one of our categories, and
initialize with random weights.
3. Continue train on the new dataset
a.) Finetune - update all parameters
b.) Freeze - update parameters only of the new FC layer - suggested
if not enough new data is available
Transfer learning effectiveness --- correct answer ---Works well if:
>> Source data is large, but target data is pretty small
>> Generalizes across tasks (object recognition params can be used
for object detection)
Limitations
>> Target data/task is completely different (silhoutte, contour)
>> Target data is large --> the random initialization is better
Power law region --- correct answer ---If data set SIZE increases in
LOG scale
than generalization ERROR decreases LINEARLY in LOG scale
Gradient based image optimization --- correct answer ---1. ) Start
with random/zero image
, 2. ) Add to the input image (I) the gradient w.r.t the score of a class
(Sc) x learning rate
3. ) Regularization (???)
Adversarial images --- correct answer ---Images, on which
gradient-based optimization was performed, but on incorrect class.
This small change fools the network, but the picture still looks like
the original class for humans. (example image about panda)
Can a change of a single pixel change the entire class prediction? ---
correct answer ---Yes
How to make NN robust towards image attacks? --- correct answer
---Include the training set images:
>> adversarial example images,
>> perturbation or
>> noise
Texture vs Shape bias:
compare humans and CNNs --- correct answer ---In image
recognition, humans have shape bias, CNNs have texture bias.
Texture vs Shape bias:
with 100% Verified Answers
INTRODUCTION:
This document contains the certified exam questions and verified
answers for CS7643 Quiz 2 (2025). It covers deep learning
architectures, CNNs, optimization, error analysis, transfer learning,
adversarial images, convolutional operations, segmentation models
(LeNet, AlexNet, VGG, ResNet, U-Net), and object detection methods
(YOLO, RPN, Mask R-CNN). The file also includes an extended study
guide with 60+ extra Q&As on neural networks, optimization
techniques, regularization, RNNs, LSTMs, GANs, reinforcement
learning, and SVMs.
EXAMS QUESTIONS AND ANSWERS:
Evolving architecture --- correct answer ---Architecture is the
subject to learn
>> evolutionary or reinforcement learning architectures
>> pruning non efficient connections
>> non-easy to optimize
,EfficientNet (ENet) --- correct answer ---CNN architecture that
operate with less parameters but perform very well.
Optimization error --- correct answer ---The model can not find the
best weights.
Higher for more complex NN with many parameters.
>> residual NN
>> evolving architectures
Estimation error --- correct answer ---Not proper generalization or
the model overfits.
Higher for more complex models.
>> regularization
>> dropout layer
Modeling error --- correct answer ---Not learning the pattern in the
dataset.
Higher for very simple NN.
--> increase the capacity
--> create more complex model
Transfer learning --- correct answer ---Reuse the features on a new
dataset, that were learned previously on a large-scale dataset.
,1. Train features on a large-scale dataset
2. Replace the last FC layer with one of our categories, and
initialize with random weights.
3. Continue train on the new dataset
a.) Finetune - update all parameters
b.) Freeze - update parameters only of the new FC layer - suggested
if not enough new data is available
Transfer learning effectiveness --- correct answer ---Works well if:
>> Source data is large, but target data is pretty small
>> Generalizes across tasks (object recognition params can be used
for object detection)
Limitations
>> Target data/task is completely different (silhoutte, contour)
>> Target data is large --> the random initialization is better
Power law region --- correct answer ---If data set SIZE increases in
LOG scale
than generalization ERROR decreases LINEARLY in LOG scale
Gradient based image optimization --- correct answer ---1. ) Start
with random/zero image
, 2. ) Add to the input image (I) the gradient w.r.t the score of a class
(Sc) x learning rate
3. ) Regularization (???)
Adversarial images --- correct answer ---Images, on which
gradient-based optimization was performed, but on incorrect class.
This small change fools the network, but the picture still looks like
the original class for humans. (example image about panda)
Can a change of a single pixel change the entire class prediction? ---
correct answer ---Yes
How to make NN robust towards image attacks? --- correct answer
---Include the training set images:
>> adversarial example images,
>> perturbation or
>> noise
Texture vs Shape bias:
compare humans and CNNs --- correct answer ---In image
recognition, humans have shape bias, CNNs have texture bias.
Texture vs Shape bias:
