Responsible and Safe AI Systems Nptel Week 3 Answers
Are you looking for Responsible and Safe AI Systems Nptel Week 3 Answers ? All weeks solutions of this Swayam course are available here.
Table of Contents
Responsible and Safe AI Systems Nptel Week 3 Answers (July-Dec 2025)
Course link: Click here to visit course on Nptel Website
Question 1. What is the definition of “Machine Unlearning”?
a) Removing the influences of training data from a trained model
b) The ability of a machine learning model to adapt to new data
c) A technique used to compress machine learning models
d) The ability of a machine learning model to learn a variety of data
Question 2. What might be some types of information that one might want to remove from model data? (Select all that apply.)
a) Private data
b) Toxic or unsafe content
c) Accurate information
d) Model hyperparameter settings
e) Stale knowledge
Question 3. What is GDPR’s Article 17 about in the context of Machine Learning?
a) Right to be remembered
b) Right to be modified
c) Right to be distributed
d) Right to be forgotten
Question 4. What are the steps in the SISA approach?
a) Sampled, Isolated, Stopped, Aggregated
b) Sharded, Imitate, Sliced, Annotated
c) Sharded, Isolated, Sliced, Aggregated
d) Sampled, Imitate, Stopped, Annotated
Question 5. What is Membership Inference Attack (MIA) used for?
a) To train LLMs faster
b) To improve the models robustness
c) To remove accurate data
d) To classify between training and unseen data
Question 6. What is the role of differential privacy?
a) To improve model accuracy
b) To make the model forget everything
c) To make models indistinguishable with/without certain data points
d) To speed up training time
Question 7. Which benchmarks are used to evaluate unlearning in LLMs? (Select all that apply.)
a) TOFU
b) GLUE
c) WMDP
d) GUIDE
Question 8. Case: A hospital wants to share patient health data with research institutions to support public health studies. However, it must ensure that a patient’s identity cannot be inferred. To achieve this, the hospital generates and sends aggregate statistics to the institute. Which form of unlearning is this?
a) Exact unlearning
b) Unlearning via differential privacy
c) Just ask for unlearning
d) Empirical Unlearning
Question 9. Which isn’t a type of graph unlearning?
a) Node unlearning
b) Edge unlearning
c) Label unlearning
d) Node feature unlearning
Question 10. Which methods are used for Node unlearning? (Select all that apply.)
a) GraphEraser
b) GUIDE
c) Projector
d) GraphdEditor
e) GNNDelete
f) MEGU
Question 11. What are “hidden representations”?
a) The final model outputs
b) Raw training data stored in memory
c) Intermediate activation vectors captured during model execution
d) The model’s loss values during training
These are Responsible and Safe AI Systems Nptel Week 3 Answers