1. Question 1

Primary goal of an autoencoder:

• ❌ Generate new data from noise

• ❌ Classify images

• ❌ Detect anomalies

• ✅ To compress and then reconstruct data

Explanation:
Autoencoders learn compact representations and try to rebuild the original input.

---

2. Question 2

In diffusion models, the forward process:

• ❌ Dimensionality reduction

• ✅ Adding noise to data in multiple steps

• ❌ Generating data

• ❌ Clustering

Explanation:
Diffusion forward pass gradually destroys structure by adding noise.

---

3. Question 3

Common unsupervised dimensionality reduction method:

• ✅ Principal Component Analysis (PCA)

• ❌ Decision trees

• ❌ K-means

• ❌ Linear regression

Explanation:
PCA projects data into lower-dimensional space using variance.

---

4. Question 4

Purpose of the bottleneck layer:

• ❌ Classification

• ❌ Generate new samples

• ✅ Lower-dimensional representation

• ❌ Final reconstruction

Explanation:
The bottleneck compresses the data into the smallest latent space.

---

5. Question 5

Common application of diffusion models:

• ❌ Language translation

• ❌ Stock prediction

• ❌ Marketing segmentation

• ✅ Enhancing image resolution / denoising

Explanation:
Diffusion models excel at image generation and denoising.

---

6. Question 6

Primary function of GAN discriminator:

• ❌ Classify categories

• ✅ Distinguish between real and fake samples

• ❌ Generate samples

• ❌ Reduce dimensionality

Explanation:
Discriminator = binary classifier (real vs fake).

---

7. Question 7

TensorFlow-supported clustering algorithm:

• ❌ CNNs

• ✅ K-means algorithm

• ❌ Linear regression

• ❌ RNNs

Explanation:
K-means is provided in TF Addons and other libraries.

---

8. Question 8

Typical activation function in GAN discriminator output:

• ❌ Tanh

• ❌ ReLU

• ❌ Softmax

• ✅ Sigmoid

Explanation:
Sigmoid gives a probability of real vs fake (binary output).

---

9. Question 9

Why noise is added to generator input?

• ❌ Increase discriminator speed

• ✅ Simulate real-world variation

• ❌ Reduce dataset size

• ❌ Make discriminator’s job easier

Explanation:
Random noise ensures diverse sample generation.

---

10. Question 10

Main goal of training a GAN:

• ❌ Improve classification accuracy

• ❌ Cluster data

• ❌ Reduce training time

• ✅ Generate realistic synthetic data

Explanation:
GANs produce new samples that mimic real data.

---

🧾 Summary Table

---