1. Question 1

• ❌ To minimize variance

• ❌ To calculate misclassified samples

• ✅ To maximize the likelihood of the correct class

• ❌ To average squared errors

---

2. Question 2

• ❌ Minimizes misclassified samples

• ❌ Increases correct classification

• ❌ Speeds up training

• ✅ It results in a flat cost surface that can cause parameters to get stuck

---

3. Question 3

• ✅ The sigmoid function provides a smooth cost surface

• ❌ Reduces misclassified samples to zero

• ❌ Increases speed

• ❌ Easier to implement

---

4. Question 4

• ❌ Sum of squared residuals

• ✅ Negative logarithm of the likelihood

• ❌ Difference between predicted and actual

• ❌ Maximum likelihood

---

5. Question 5

• ❌ nn.BCELoss

• ❌ nn.MSELoss

• ❌ nn.SoftmaxLoss

• ✅ nn.CrossEntropyLoss

---

6. Question 6

• ❌ Faster than all optimizers

• ✅ Uses only a portion of the dataset to minimize loss

• ❌ Guarantees global minimum

• ❌ Increases learning rate automatically

---

7. Question 7

• ✅ Converts linear outputs into probabilities

• ❌ Thresholding

• ❌ Computes gradient

• ❌ Adjusts learning rate

---

8. Question 8

• ❌ Updates parameters

• ❌ Resets parameters

• ✅ Computes gradients w.r.t model parameters

• ❌ Calculates next loss

---

9. Question 9

• ❌ Averaging input data

• ✅ Using optimizer.step()

• ❌ Recalculating loss

• ❌ Increasing learning rate

---

10. Question 10

• ❌ 0–10

• ✅ 0–1

• ❌ –1 to 1

• ❌ 0–100

---

🧾 Summary Table

---