Domain: Foundations & Theory
Controlled experiment comparing variants by random assignment to estimate causal effects of changes.
Fraction of correct predictions; can be misleading on imbalanced datasets.
Nonlinear functions enabling networks to approximate complex mappings; ReLU variants dominate modern DL.
Selecting the most informative samples to label (e.g., uncertainty sampling) to reduce labeling cost.
Methods like Adam adjusting learning rates dynamically.
Inputs crafted to cause model errors or unsafe behavior, often imperceptible in vision or subtle in text.
Ensuring model behavior matches human goals, norms, and constraints, including reducing harmful or deceptive outputs.
The field of building systems that perform tasks associated with human intelligence—perception, reasoning, language, planning, and decision-making—via algori...
Scalar summary of ROC; measures ranking ability, not calibration.
Tendency to trust automated suggestions even when incorrect; mitigated by UI design, training, and checks.
Generates sequences one token at a time, conditioning on past tokens.
Hidden behavior activated by specific triggers, causing targeted mispredictions or undesired outputs.
Number of samples per gradient update; impacts compute efficiency, generalization, and stability.
Search algorithm for generation that keeps top-k partial sequences; can improve likelihood but reduce diversity.
Systematic differences in model outcomes across groups; arises from data, labels, and deployment context.
A conceptual framework describing error as the sum of systematic error (bias) and sensitivity to data (variance).
The degree to which predicted probabilities match true frequencies (e.g., 0.8 means ~80% correct).
Framework for reasoning about cause-effect relationships beyond correlation, often using structural assumptions and experiments.
Stepwise reasoning patterns that can improve multi-step tasks; often handled implicitly or summarized for safety/privacy.
Breaking documents into pieces for retrieval; chunk size/overlap strongly affect RAG quality.
Hardware resources used for training/inference; constrained by memory bandwidth, FLOPs, and parallelism.
The relationship between inputs and outputs changes over time, requiring monitoring and model updates.
A hidden variable influences both cause and effect, biasing naive estimates of causal impact.
A table summarizing classification outcomes, foundational for metrics like precision, recall, specificity.
Optimization under equality/inequality constraints.
Mathematical framework for controlling dynamic systems.
Algorithm computing control actions.
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.