---
source_file: "Deep Learning - UTS - 2023 - CSA.pdf.pdf"
source_type: "scanned PDF (1 page, image-only — no embedded text layer)"
transcription_method: "page rasterization (pypdfium2) + visual reading"
course: "Deep Learning [IUP]"
exam: "Mid Term Examination (UTS) — Midterm"
term: "Semester I Academic Year 2023/2024"
program: "Undergraduate Program in Computer Science (IUP)"
faculty: "Faculty of Mathematics and Natural Sciences (FMIPA)"
university: "Universitas Gadjah Mada (UGM)"
date: "Monday, October 09, 2023"
time: "07:30–09:30 (120 menit)"
lecturer: "Wahyono, Ph.D."
exam_type: "Open Book"
pages: 1
topics: [hyperparameters, multi-layer-perceptron, perceptron-training, backpropagation, activation-function, overfitting]
---

# UTS Deep Learning [IUP] — 2023/2024

> **Header (as printed):**
> UNDERGRADUATE PROGRAM IN COMPUTER SCIENCE
> FACULTY OF MATHEMATICS AND NATURAL SCIENCES
> UNIVERSITAS GADJAH MADA
>
> Mid Term Examination
> Semester I Academic Year 2023/2024

| Field | Value |
|---|---|
| Course Name | Deep Learning [IUP] |
| Date | Monday, October 09, 2023 |
| Time | 07:30–09:30 (120 menit) |
| Lecturer | Wahyono, Ph.D. |
| Exam Types | Open Book |

---

## CO1 – PLO2: *Able to differentiate Deep Learning to Traditional Neural Network (7.5%);*

### 1. Answer the questions below:

a. Explain the effect of using hyperparameters in multilayer perceptron!

b. Explain why although they have the same number of layers, deep learning always produces better accuracy than multilayer perceptron!

---

## CO2 – PLO2: *Able to understand multi-layer perceptron and backpropagation (10%);*

### 2. Given the following data:

| X1 | X2 | X3 | Y |
|---|---|---|---|
| 1 | 1 | -1 | 1 |
| 1 | 1 | 1 | 1 |
| -1 | 1 | 1 | -1 |
| -1 | -1 | 1 | -1 |

Using Single Layer Perceptron, determine the values of W0, W1, W2, and W3 after second epoch if following parameters are known!

a. Learning rate=0.2, zero weight initialization, and stochastic gradient descent optimizer.

b. Learning rate=0.3, zero weight initialization, and batch gradient descent optimizer.

---

## CO3 – PLO2: *Able to understand components in DNN architecture. (7.5%);*

### 3. Why is an activation function needed in the training process with deep learning? Explain the benefits of activation function?

### 4. One of the common problems in the deep neural network is overfitting. Explain the overfitting condition and 4 strategies to handle overfitting!

---

-- Good Luck --

*(Page 1)*