Machine Learning Deep Learning 5

Question 1

What is the primary advantage of using a pre-trained word embedding like Word2Vec or GloVe?

Accepted Answer

They provide dense semantic representations that capture word relationships from large corpora

Answer

Pretrained embeddings encode semantic similarity (e.g., king − man + woman ≈ queen) learned from billions of words, giving models a strong starting point.

Question 2

In a U-Net architecture, what is the purpose of the skip connections between encoder and decoder?

Accepted Answer

To pass high-resolution feature maps from the encoder to corresponding decoder layers

Answer

U-Net skip connections concatenate encoder feature maps to decoder inputs, preserving spatial detail lost during downsampling for precise pixel-level predictions.

Question 3

What is model quantization in the context of deploying deep learning models?

Accepted Answer

Reducing the numerical precision of weights (e.g., float32 → int8) to speed up inference

Answer

Quantization lowers the bit-width of weights and activations, reducing memory footprint and latency at the cost of a small accuracy drop.

Question 4

What distinguishes a variational autoencoder (VAE) from a standard autoencoder?

Accepted Answer

VAE learns a probabilistic latent space and can generate new samples by sampling from it

Answer

VAE encodes inputs as distributions (mean and variance) over a latent space, enabling smooth interpolation and generation of new data by sampling.

Question 5

What is knowledge distillation in deep learning?

Accepted Answer

Training a smaller student model to mimic the soft outputs of a larger teacher model

Answer

Knowledge distillation trains a compact student network to match the probability outputs of a larger teacher, capturing the teacher's learned knowledge efficiently.

Machine Learning Practice Test

Machine Learning Practice Test

Machine Learning Deep Learning 5