[STUDY GUIDE] • HIGH-YIELD PRACTICE & REVIEW EDITION
CS 7643 Quiz 3 Questions & Answer Key
Solutions Latest 2026/2027 Edition
Comprehensive Examination Question Bank • Concept Mapping
TOTAL QUESTIONS EXAM TOPICS ANSWERS
25 Questions 10 Modules 100% Verified
DOCUMENT OVERVIEW
This document contains 25 verified questions with correct answers focusing on advanced concepts in deep learning, particularly
recurrent neural networks and word embeddings. Each question is paired with a concise answer and explanation, making it
suitable for students preparing for exams, undertaking coursework, or seeking certification in machine learning and natural
language processing.
EXAM BLUEPRINT & TOPIC DISTRIBUTION
Systematic breakdown of subject domains and exam coverage.
Topic Module Scope & Core Focus Questions Share (%)
Explores the principles and differentiating factors of RNNs and
Recurrent Neural Networks LSTMs. 4 Qs 16.0%
Examines various methods and evaluations related to word
Word Embeddings embeddings. 6 Qs 24.0%
Discusses the various types and applications of conditional language
Conditional Language Models models. 2 Qs 8.0%
Analyzes the concept and implementation of negative sampling in
Negative Sampling machine learning. 3 Qs 12.0%
t-SNE and Dimensionality Focuses on t-SNE as a technique for visualizing high-dimensional
Reduction data. 3 Qs 12.0%
Covers the transfer of knowledge from larger models to smaller
Knowledge Distillation models. 2 Qs 8.0%
Introduces various embedding techniques beyond traditional word
Graph and Other Embeddings embeddings. 2 Qs 8.0%
Confidential • Student Study Edition • Practice & Review Guide Page 1 of 10
, STUDENT STUDY & MASTERY EDITION PRACTICE & REVIEW GUIDE
Evaluation Methods for Details intrinsic and extrinsic evaluation methods for assessing
Embeddings embeddings. 1 Qs 4.0%
Discusses the Word2Vec algorithm and its relevance in machine
Word2Vec and Algorithms learning. 1 Qs 4.0%
Explains how word meanings are derived from their distributional
Distribution Semantics properties. 1 Qs 4.0%
Total Exam Coverage 10 Integrated Topic Modules 25 Qs 100.0%
TOPIC 1: RECURRENT NEURAL NETWORKS
4 Questions • 16.0% of Exam • Explores the principles and differentiating factors of RNNs and LSTMs.
QUESTION 1
How do RNN and LSTM update rules differ?
Answer: LSTM networks update rule is cell state is updated in an additive way by adding something to its previous value
C_t-1, this differs from the multiplicative update Rule of RNN.
QUESTION 2
Gradients for RNNs
Figure Q2: Reference Image
Answer: The gradient computation involves recurrent multiplication of WW. This multiplying by WW to each cell has a
bad effect. Think like this: If you a scalar (number) and you multiply gradients by it over and over again for say 100 times,
if that number > 1, it'll explode the gradient and if < 1, it'll vanish towards 0.
Confidential • Student Study Edition • Practice & Review Guide Page 2 of 10
CS 7643 Quiz 3 Questions & Answer Key
Solutions Latest 2026/2027 Edition
Comprehensive Examination Question Bank • Concept Mapping
TOTAL QUESTIONS EXAM TOPICS ANSWERS
25 Questions 10 Modules 100% Verified
DOCUMENT OVERVIEW
This document contains 25 verified questions with correct answers focusing on advanced concepts in deep learning, particularly
recurrent neural networks and word embeddings. Each question is paired with a concise answer and explanation, making it
suitable for students preparing for exams, undertaking coursework, or seeking certification in machine learning and natural
language processing.
EXAM BLUEPRINT & TOPIC DISTRIBUTION
Systematic breakdown of subject domains and exam coverage.
Topic Module Scope & Core Focus Questions Share (%)
Explores the principles and differentiating factors of RNNs and
Recurrent Neural Networks LSTMs. 4 Qs 16.0%
Examines various methods and evaluations related to word
Word Embeddings embeddings. 6 Qs 24.0%
Discusses the various types and applications of conditional language
Conditional Language Models models. 2 Qs 8.0%
Analyzes the concept and implementation of negative sampling in
Negative Sampling machine learning. 3 Qs 12.0%
t-SNE and Dimensionality Focuses on t-SNE as a technique for visualizing high-dimensional
Reduction data. 3 Qs 12.0%
Covers the transfer of knowledge from larger models to smaller
Knowledge Distillation models. 2 Qs 8.0%
Introduces various embedding techniques beyond traditional word
Graph and Other Embeddings embeddings. 2 Qs 8.0%
Confidential • Student Study Edition • Practice & Review Guide Page 1 of 10
, STUDENT STUDY & MASTERY EDITION PRACTICE & REVIEW GUIDE
Evaluation Methods for Details intrinsic and extrinsic evaluation methods for assessing
Embeddings embeddings. 1 Qs 4.0%
Discusses the Word2Vec algorithm and its relevance in machine
Word2Vec and Algorithms learning. 1 Qs 4.0%
Explains how word meanings are derived from their distributional
Distribution Semantics properties. 1 Qs 4.0%
Total Exam Coverage 10 Integrated Topic Modules 25 Qs 100.0%
TOPIC 1: RECURRENT NEURAL NETWORKS
4 Questions • 16.0% of Exam • Explores the principles and differentiating factors of RNNs and LSTMs.
QUESTION 1
How do RNN and LSTM update rules differ?
Answer: LSTM networks update rule is cell state is updated in an additive way by adding something to its previous value
C_t-1, this differs from the multiplicative update Rule of RNN.
QUESTION 2
Gradients for RNNs
Figure Q2: Reference Image
Answer: The gradient computation involves recurrent multiplication of WW. This multiplying by WW to each cell has a
bad effect. Think like this: If you a scalar (number) and you multiply gradients by it over and over again for say 100 times,
if that number > 1, it'll explode the gradient and if < 1, it'll vanish towards 0.
Confidential • Student Study Edition • Practice & Review Guide Page 2 of 10
