10-601 Machine Learning Exam 3 Practice Problems - Solutions

Aug 30, 2026

Uploaded by Margaret

Exam 3 practice solutions for a 10-601 machine learning course, updated April 23, 2024. The material answers multiple CNN-focused questions, covering what a convolutional kernel is and how stride and padding affect output shape and information preservation. It derives output-size formulas, suggests follow-up layers that produce a 1×1 binary label, and compares upsampling versus downsampling and their appropriate use cases. It also discusses parameter sharing and counts learnable parameters.

LanguageEnglish

CategoryExam

FormatPDF

Pages35

Reading Time1 h 28 min

AccessFree to download

Instructions

Exam rules and response formatting

Select One questions

Select all that apply questions

Problem 1: CNN kernel components

What is a kernel?

Why stride matters

Role and benefits of padding

Problem 2: Output shapes and follow-up classification

Smallest stride/padding for a target output

Deriving general output size b

Producing a 1×1 binary label

Problem 3: Upsampling vs downsampling

Definitions and scenarios

Problem 4: Parameter sharing

Parameter count with and without sharing

Solutions 10-601 Machine Learning Name: Spring 2024 AndrewID: Exam 3 Practice Problems Updated: April 23, 2024 Time Limit: N/A Instructions: • Fill in your name and Andrew ID above. Be sure to write neatly, or you may not receive credit for your exam. • Clearly mark your answers in the allocated space on the front of each page. If needed, use the back of a page for scratch space, but you will not get credit for anything written on the back of a page. If you have made a mistake, cross out the invalid parts of your solution, and circle the ones which should be graded. • No electronic devices may be used during the exam. • Please write all answers in pen. • You have N/A to complete the exam. Good luck! Instructions for Specific Problem Types For “Select One” questions, please fill in the appropriate bubble completely: Select One: Who taught this course? Henry Chai \# Marie Curie \# Noam Chomsky If you need to change your answer, you may cross out the previous answer and bubble in the new answer: Select One: Who taught this course? Henry Chai \# Marie Curie Noam Chomsky For “Select all that apply” questions, please fill in all appropriate squares completely: Select all that apply: Which are scientists? ■ Stephen Hawking ■ Albert Einstein ■ Isaac Newton □ I don’t know Again, if you need to change your answer, you may cross out the previous answer(s) and bubble in the new answer(s): Select all that apply: Which are scientists? ■ Stephen Hawking ■ Albert Einstein ■ Isaac Newton I don’t know For questions where you must fill in a blank, please make sure your final answer is fully included in the given space. You may cross out answers or parts of answers, but the final answer must still be within the given space. Fill in the blank: What is the course number? 10-601 10-601 1 CNNs and RNNs 1. Let’s begin by considering some of the high-level components of a CNN kernel along with the basic motivation. i. What is a kernel? The collection of weights/filters that you will pass through an image, along with hyperparameters such as padding, stride, and filter size/quantity. ii. Why do we need stride, and what benefits/tradeoffs might different values of stride have on the output? Stride is what allows the filter to actually pass through the CNN (defines at what pace the filter will move across the rows/columns) . Larger values of stride can allow you to reduce the output dimensionality which could combat overfitting along with reducing computational power. The downside however is that you lose more and more information with larger values of stride, which could limit the upside of your model’s accuracy. iii. What functionality does padding add to the kernel? Why might we want to use it? | | | --- | | | | | Padding essentially makes sure that the output shape is the same as the input shape, and allows every pixel to be included in the convolution: if our filter is about to slide off of the original image padding allows the filter/image to still line up correctly. Furthermore, padding helps filters focus on the corner pixels just as much as middle pixels by making the filter pass through the corners multiple times as opposed to just once. 2. Consider the following image, filter, and output shape, which you have seen in prior homework in this course. The shape of this particular Y is that of an kernel using no padding and a stride of 1 . i. Suppose we decide that, instead of having our output shape be (4 , 4), we want a slightly smaller, (3 , 3) image as output for the kernel. In order for this to happen, what is the smallest combination of stride and padding that would work? s=2, p=1 ii. Let’s make this a bit more general. Suppose our original image of shape (a, a), and we want the shape of our final image to be of shape (b, b), where b ≤ a. Furthermore the shape of the filter is (k, k) and the stride length is s. Express b in terms of all

FAQs

What is a CNN kernel, and what elements does it include?

A kernel is the collection of weights/filters applied to an image, together with hyperparameters such as padding, stride, and filter size/quantity.

How does stride affect CNN output and model behavior?

Stride controls how quickly the filter moves across the input. Larger stride can reduce output dimensionality and computation, which may help mitigate overfitting, but it can lose more information and limit accuracy gains.

When should upsampling or downsampling be used in CNNs?

Downsampling reduces output dimensionality, commonly via pooling, and is useful for larger images to reduce compute and complexity. Upsampling increases output dimensionality and is useful when the output needs to be larger or matched to the input, such as per-pixel labeling.

Aug 30, 2026

Uploaded by Margaret