Scribd
Upload
0 views2 pages

MTech (Module 01 QP)-2

The document outlines a series of questions related to machine learning concepts, including bias, maximum likelihood estimation, learning algorithms, hyperparameters, and the differences between supervised and unsupervised learning. It also addresses more complex topics such as overfitting, the bias-variance tradeoff, and the design and evaluation of machine learning systems. The questions are categorized into 4-mark, 6-mark, and 10-mark sections, indicating varying levels of depth and analysis required for responses.

Uploaded by

Mohith Kalyan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0 views2 pages

MTech (Module 01 QP)-2

The document outlines a series of questions related to machine learning concepts, including bias, maximum likelihood estimation, learning algorithms, hyperparameters, and the differences between supervised and unsupervised learning. It also addresses more complex topics such as overfitting, the bias-variance tradeoff, and the design and evaluation of machine learning systems. The questions are categorized into 4-mark, 6-mark, and 10-mark sections, indicating varying levels of depth and analysis required for responses.

Uploaded by

Mohith Kalyan
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

4-Mark Questions

1.​ Define the concept of bias in machine learning. How does it impact the performance of a
model?
2.​ State the objective of Maximum Likelihood Estimation (MLE).
3.​ Define a learning algorithm with an example.
4.​ What is the difference between capacity and effective capacity?
5.​ What are hyperparameters? Give examples.
6.​ Differentiate between supervised and unsupervised learning.
7.​ Define validation set and explain its use.
8.​ State the objective of Maximum Likelihood Estimation (MLE).
9.​ What is the primary difference between frequentist and Bayesian approaches?
10.​Write a short note on stochastic gradient descent (SGD).

6-Mark Questions

1.​ Discuss the problem of overfitting with a suitable diagram and example.
2.​ Describe the steps involved in building a machine learning algorithm. Use the linear
regression example as a reference.
3.​ Discuss the problem of overfitting with a suitable diagram and example.
4.​ Explain the bias-variance tradeoff and its implications on model performance.
5.​ Describe how hyperparameter tuning is performed using validation sets.
6.​ Explain the principle of maximum likelihood estimation with a simple likelihood function.
7.​ Compare and contrast supervised and unsupervised learning algorithms with examples.
8.​ Discuss the need for using a test set in evaluating performance.
9.​ Describe the challenges in choosing a good performance measure.
10.​Explain the use of SGD in training machine learning models, including its benefits.
11.​You are designing a machine learning model to classify emails as “spam” or “not spam.”
You have access to a dataset of 50,000 labeled emails. Explain how you would define
the task T, performance measure P, and experience E for this system. Also, discuss what
type of learning (supervised/unsupervised) this system represents.
12.​You are working on a real estate project where your model predicts house prices based
on input features such as number of bedrooms, location, and square footage. Explain
how you would define the task T, performance measure P, and experience E for this
system. Which performance metric would you use and why? Would classification be
suitable here?

10-Mark Questions (Long/Analytical)


1.​ You are designing a machine learning model to classify emails as “spam” or “not spam.”​
Define the task (T), performance measure (P), and experience (E). Also, explain what
type of learning this is (supervised or unsupervised) and how you would evaluate the
model’s effectiveness.
2.​ Analyze the concept of bias and variance in the context of model evaluation. Explain
how the bias–variance trade-off affects generalization, and illustrate your answer with an
error curve showing training and test performance.
3.​ With appropriate examples, explain the concepts of capacity, underfitting, and overfitting.
How do these relate to the training and test error?
4.​ Describe the steps involved in building a machine learning algorithm. Use the linear
regression example as a reference.
5.​ Explain estimators, bias, variance, and their relationship in evaluating model
performance.
6.​ Compare Maximum Likelihood Estimation and Bayesian Estimation in detail. Discuss
their advantages and drawbacks.
7.​ Describe the complete process of designing and evaluating a machine learning system,
including task definition, performance measures, and experience.
8.​ Explain in depth the challenges that motivated the emergence of deep learning
approaches over traditional machine learning techniques.
9.​ You are developing a machine learning system that transcribes spoken audio into text.
You have a dataset of audio clips paired with corresponding transcripts.(a) Define the
type of learning involved and identify the task T, experience E, and performance
measure P. (b) What challenges would you face in choosing the performance measure?
Should you score entire sequence correctness or partial accuracy? Justify your answer.
10.​A bank wants to use machine learning to detect fraudulent transactions. However,
fraudulent cases are rare in the dataset.(a) How would you design the task, experience,
and performance measure for this use case?(b) Why would accuracy be a misleading
metric in this context? Suggest and justify a more appropriate metric.
11.​9. You have a dataset of 100,000 images, but only 2,000 of them are labeled. You are
asked to build an image classification model.(a) What learning approach would you
choose (supervised, semi-supervised, or unsupervised)? Why? (b) What strategies can
you use to improve model performance with such limited labeled data?

You might also like