Computer — All MCQs
4795 questions available
Q2041.Which AI risk involves crafting inputs to deliberately fool a model?
Correct Answer: D
Adversarial attacks use specially crafted inputs to deceive AI models.
Q2042.Which technique compresses a large model into a smaller one while retaining performance?
Correct Answer: A
Knowledge distillation transfers knowledge from a large 'teacher' to a small 'student' model.
Q2043.An AI 'context window' refers to:
Correct Answer: A
The context window is the maximum amount of tokens an LLM processes in one prompt.
Q2044.Which is a prime concern of the EU AI Act?
Correct Answer: A
The EU AI Act regulates AI using a risk-based classification framework.
Q2045.Which AI method is best to predict the next likely location of repeat offenders?
Correct Answer: D
Predictive analytics models patterns to forecast likely crime hotspots.
Q2046.Which embedding model captures word meaning using surrounding context (contextual embeddings)?
Correct Answer: B
BERT produces contextual embeddings that vary with surrounding words.
Q2047.In a CNN, the 'kernel/filter' is responsible for:
Correct Answer: D
Convolutional kernels slide over images to detect local features such as edges.
Q2048.Which approach trains a model across decentralized devices without sharing raw data?
Correct Answer: B
Federated learning trains models on-device, sharing updates not raw private data.
Q2049.Which AI concept is central to detecting previously unseen malware (zero-day threats)?
Correct Answer: D
Behavior-based anomaly detection can catch novel threats lacking known signatures.
Q2050.The 'temperature' parameter in an LLM controls:
Correct Answer: C
Higher temperature increases output randomness; lower makes outputs more deterministic.
Q2051.Which is a limitation of rule-based expert systems compared to ML?
Correct Answer: B
Rule-based systems require manual rules and handle uncertainty/scale poorly.
Q2052.Which AI technique would help transcribe and translate intercepted foreign-language calls?
Correct Answer: B
ASR transcribes speech and machine translation converts it across languages.
Q2053.What does 'precision' measure in classification?
Correct Answer: D
Precision = TP / (TP + FP): correctness among predicted positives.
Q2054.What does 'recall' measure?
Correct Answer: D
Recall = TP / (TP + FN): proportion of actual positives correctly identified.
Q2055.Which AI governance principle requires that individuals can contest automated decisions?
Correct Answer: C
Governance frameworks grant rights to explanation and to contest automated decisions.
Q2056.Which method handles class imbalance by generating synthetic minority samples?
Correct Answer: B
SMOTE (Synthetic Minority Over-sampling Technique) creates synthetic minority-class samples.
Q2057.An investigator uses AI to cluster similar cybercrime cases by modus operandi.
Correct Answer: D
Grouping cases by similarity without labels is unsupervised clustering.
Q2058.Which is TRUE about 'gradient descent'?
Correct Answer: A
Gradient descent iteratively updates parameters to minimize the loss function.
Q2059.What distinguishes 'stochastic gradient descent' (SGD) from batch gradient descent?
Correct Answer: B
SGD updates weights using one or a few samples per step, speeding training.
Q2060.Which AI risk describes a model leaking private training data through its outputs?
Correct Answer: B
Models can memorize and leak sensitive training data through their outputs.