Deep Learning (Data Scientist) 4

Question 1

A data scientist deploys a fraud-detection network trained on 2022 data and observes accuracy decay in production during 2024. What is this phenomenon called?

Accepted Answer

Data drift (distribution shift)

Answer

When the production data distribution diverges from the training distribution over time, model performance degrades due to data drift.

Question 2

In a GAN, the generator produces nearly identical outputs regardless of the input noise vector. What is this failure mode called?

Accepted Answer

Mode collapse

Answer

Mode collapse occurs when the generator maps diverse noise inputs to a narrow set of outputs, ignoring the data's variety.

Question 3

Which practice most directly reduces the memory footprint and latency of a trained network for edge deployment?

Accepted Answer

Quantization of weights to lower-precision formats such as INT8

Answer

Quantization stores and computes weights in lower precision, shrinking model size and speeding up inference on edge hardware.

Question 4

A team wants uncertainty estimates from a deployed classifier. Which approach provides an approximate Bayesian estimate at inference time?

Accepted Answer

Monte Carlo dropout, keeping dropout active during multiple forward passes

Answer

Running several stochastic forward passes with dropout enabled approximates sampling from a posterior, yielding predictive uncertainty.

Question 5

Why is ReLU generally preferred over sigmoid for hidden layers in deep networks?

Accepted Answer

ReLU avoids saturation for positive inputs, preserving gradient magnitude in deep stacks

Answer

Sigmoid saturates at both extremes and shrinks gradients, while ReLU's linear positive region keeps gradients flowing in deep networks.

Deep Learning Practice Test

Deep Learning Practice Test

Deep Learning (Data Scientist) 4