Difficulty: Intermediate
Predicting borrower default risk.
Attention between different modalities.
Measures divergence between true and predicted probability distributions.
A robust evaluation technique that trains/evaluates across multiple splits to estimate performance variability.
Ordering training samples from easier to harder to improve convergence or generalization.
Expanding training data via transformations (flips, noise, paraphrases) to improve robustness.
Shift in feature distribution over time.
Processes and controls for data quality, access, lineage, retention, and compliance across the AI lifecycle.
Human or automated process of assigning targets; quality, consistency, and guidelines matter heavily.
When information from evaluation data improperly influences training, inflating reported performance.
Tracking where data came from and how it was transformed; key for debugging and compliance.
Maliciously inserting or altering training data to implant backdoors or degrade performance.
Privacy risk analysis under GDPR-like laws.
Increasing performance via more data.
A structured collection of examples used to train/evaluate models; quality, bias, and coverage often dominate outcomes.
Differences between training and deployed patient populations.
Structured dataset documentation covering collection, composition, recommended uses, biases, and maintenance.
A branch of ML using multi-layer neural networks to learn hierarchical representations, often excelling in vision, speech, and language.
Tradeoffs between many layers vs many neurons per layer.
A formal privacy framework ensuring outputs do not reveal much about any single individual’s data contribution.
Accelerating safety relative to capabilities.
Training a smaller “student” model to mimic a larger “teacher,” often improving efficiency while retaining performance.
Train/test environment mismatch.
A mismatch between training and deployment data distributions that can degrade model performance.
A preference-based training method optimizing policies directly from pairwise comparisons without explicit RL loops.
Randomly zeroing activations during training to reduce co-adaptation and overfitting.
Alternative formulation providing bounds.
Legal right to fair treatment.
AI-assisted review of legal documents.
Halting training when validation performance stops improving to reduce overfitting.