Natural Language Processing
CSE · Final · Fall 24
Paper 2 of 2Section Re_ASwitch
Section Re_A
Paper 2 of 2
Read the questions as text
Read off the PDF by AI, so it may have mistakes. The PDF is the original.
Daffodil International University
Department of Computer Science and Engineering
Faculty of Science & Information Technology
Final Examination, Fall - 2024
Course Code: CSE445, Course Title: Natural Language Processing
Section: Re_A
Time: 2 Hours
Marks: 40
Answer All Questions
[The figures in the right margin indicate the full marks and corresponding course outcomes. All portions of each question must be answered sequentially.]
1. Explain one-hot encoding as a method for representing words as vectors. Discuss its limitations why this poses a challenge in natural language processing. | 5 | CO2
2. (a) Explain the concept of Latent Dirichlet Allocation (LDA) in topic modeling. | 2 | 3
(b) Describe how Collapsed Gibbs Sampling (CGS) is applied in LDA to infer topic distributions with example. Create 2 topics on your own to illustrate the CGS.
3. Imagine you're typing on your phone, and it suggests the next word based on what you've already written. A recurrent network is working behind this suggestion. It processes sequential data one step at a time, maintaining a hidden state that captures information about previous inputs to influence future outputs. Now, your task is to draw that model's architecture and explain its advantages and disadvantages. | 5
4. (a) Define and analyze 3 types of activation function (sigmoid, ReLu, Tanh) with graphical representation and required equations. | 5
(b) Assume that the neuron has a sigmoid activation function, apply a Backpropagation algorithm to perform a forward pass and a backward pass to update the value of w35 and w45 on the network. Assume that the learning rate is 1. | 10
[Figure: A neural network diagram with input nodes X1 = 0.35, X2 = 0.9, hidden nodes H3 with bias b1 = 0.1, H4, and output node O5 with bias b2 = 0.25, target Y5 = 0.1, and weights w13 = 0.2, w14 = 0.1, w23 = 0.8, w24 = 0.7, w35 = 0.4, w45 = 0.11]
5. A team is working on developing a speech recognition system for a voice-activated assistant. To ensure the system performs accurately, they need to create a robust dataset, preprocess the speech data, and implement appropriate analysis techniques.
(a) Explain two preprocessing techniques that can be applied to the speech data to improve its quality before analysis. Provide examples of tools or methods that could be used. | 4 | CO2, CO3
(b) Write down the importance of speech dataset for the system is being developed and the possible challenges that they may face while collecting data. | 6
Best of luck!