CNN Basics
What does the abbreviation CNN stand for?
Convolutional Neural Network
Which task involves both locating an object in an image and classifying it?
Object Detection
Question: Which of these is NOT a standard layer in a CNN?
A) Social Layer B) Pooling Layer C) Convolutional Layer D) Fully Connected Layer
A) Social Layer
Which programming language is the most widely used for Machine Learning today?
Python
In the TV show Silicon Valley, an app was created that could only classify two things in the world. Name them.
Hotdog and Not Hotdog.
Which activation function replaces all negative pixel values with zero?
ReLU (Rectified Linear Unit)
What is the name of the famous real-time object detection algorithm that stands for "You Only Look Once"?
YOLO
Question: Which of these is NOT a Computer Vision task?
A) Image Segmentation B) Object Detection C) Audio Summarization D) Edge Detection
C) Audio Summarization
Which hardware component is essential for training deep neural networks due to its parallel processing power?
GPU (Graphics Processing Unit)
If you have an infinite amount of data and an infinite amount of computing power, do you still need to worry about Overfitting?
No
What term describes the number of pixels by which a filter shifts over the input matrix?
Stride
What is the process of partitioning an image into multiple segments (sets of pixels) to identify boundaries?
Image Segmentation
Question: Which of these is NOT a popular CNN architecture?
A) YOLO B) MobileNet C) ResNet D) SkyNet
D) SkyNet
Which type of learning uses a dataset that already contains the correct answers or "labels"?
Supervised Learning
Our machine learning instructor's full name, including patronymic.
Bazarkhanova Aigerim Adilzhankyzy
What do we call the process of adding extra zero-pixels around the border of an image?
Padding
Which technique is used to identify the sharp changes in brightness to find the boundaries of objects?
Edge Detection
Question: Which of these is NOT an activation function?
A) ReLU B) Backprop C) Sigmoid D) Softmax
B) Backprop (It's an algorithm, not a function)
We usually divide our data into two parts: one to teach the model and one to check how well it learned. Name these two sets.
Training set and Test set
If an AI can hold a conversation so well that you can't tell it's a machine, which famous 1950s test has it passed?
The Turing Test.
What type of layer is used to reduce the spatial size (dimension) of the representation to reduce the number of parameters?
Pooling layer (e.g., Max Pooling)
What is the term for identifying significant patterns like textures, shapes, or corners in an image?
Feature Extraction
Which of these is NOT a Foundation Model / SOTA mentioned in the lecture?
D) DeepBlue (It's the chess computer from the 90s)
In ML, what is the general term for the difference between the model's prediction and the actual correct answer? (Hint: The model tries to minimize this).
Error or Loss
If you have 1,000 monkeys typing on 1,000 laptops for a million years, they might write Shakespeare. But if you have 1 GPU and a week, you can train a model to do it. What do we call this field?
Generative AI (or LLMs).