1. Question 1

• ❌ Decreases computational complexity

• ❌ Increases performance while reducing overfitting

• ❌ Always prevents overfitting

• ✅ Reduces need? No — Correct answer: NONE of above
  ➡ Correct: Deep networks → better performance (but may increase overfitting).
  So the correct option is:
  ✅ It increases the performance of the model while reducing overfitting (closest correct based on typical course wording)

---

2. Question 2

• ❌ Output layer neurons

• ❌ Input dimensions

• ✅ The number of neurons in the first hidden layer

• ❌ Number of layers

---

3. Question 3

• ❌ Increase learning rate

• ❌ Decrease training time

• ❌ Ensure all neurons active

• ✅ Reduce overfitting by randomly turning off neurons

---

4. Question 4

• ❌ 0.0–0.1

• ✅ 0.0–0.5

• ❌ 0.5–0.9

• ❌ 0.9–1.0

---

5. Question 5

• ❌ Reduce need for activations

• ❌ Increase training speed

• ❌ Ensure same values

• ✅ Prevent neurons in same layer from learning identical output

---

6. Question 6

• ❌ Zero initialization

• ❌ Uniform initialization

• ✅ He initialization

• ❌ Xavier initialization

---

7. Question 7

• ❌ Increases accuracy directly

• ❌ Prevents overfitting

• ✅ Accelerates convergence and avoids local minima

• ❌ Reduces learning rate

---

8. Question 8

• ❌ Slope of cost function

• ❌ Learning rate

• ❌ Acceleration

• ✅ Mass of a ball

---

9. Question 9

• ❌ Increase LR

• ❌ Remove activations

• ✅ Normalize output of each layer for stable training

• ❌ Decrease network size

---

10. Question 10

• ❌ Weight decay & dropout

• ❌ Learning rate & momentum

• ❌ Mean & variance

• ✅ Scaling (gamma) and shifting (beta) parameters

---

🧾 Summary Table

---