Deep Learning (Data Scientist) 5

Question 1

A data scientist fine-tunes a large language model on task B and finds it now performs poorly on the original task A. What is this called?

Accepted Answer

Catastrophic forgetting

Answer

Mode collapse

Answer

Double descent

Answer

Gradient checkpointing

Question 2

Which initialization scheme is specifically designed for layers with ReLU activations?

Accepted Answer

He (Kaiming) initialization

Answer

Zero initialization

Answer

Xavier initialization tuned for tanh

Answer

Identity initialization

Question 3

Class imbalance is severe: 99% negative, 1% positive. Which combination best supports training a useful classifier?

Accepted Answer

Class-weighted loss or resampling, evaluated with precision-recall metrics

Answer

Standard cross-entropy evaluated with accuracy

Answer

Removing all negative examples

Answer

Training longer with the same setup

Question 4

What is the primary benefit of mixed-precision (FP16/FP32) training on modern GPUs?

Accepted Answer

Faster computation and lower memory use with minimal accuracy loss

Answer

Guaranteed elimination of overfitting

Answer

Removal of the need for a loss function

Answer

Automatic hyperparameter selection

Question 5

In self-supervised learning, how does contrastive pretraining like SimCLR create labels?

Accepted Answer

It treats augmented views of the same image as positives and other images as negatives

Answer

It uses crowdsourced human annotations

Answer

It copies labels from a supervised teacher model

Answer

It clusters images by file size

Question 6

A learning-rate schedule that starts small, increases briefly, then decays following a cosine curve is using which technique?

Accepted Answer

Warmup followed by cosine annealing

Answer

Exponential gradient clipping

Answer

Nesterov lookahead

Answer

Layer-wise freezing

Question 7

Which method explains an individual prediction of a deep model by attributing importance scores to input features?

Accepted Answer

SHAP or integrated gradients

Answer

Batch normalization

Answer

Beam search

Answer

Stochastic weight averaging