Difficulty: Intermediate
Using production outcomes to improve models.
Model trained on its own outputs degrades quality.
Achieving task performance by providing a small number of examples inside the prompt without weight updates.
Updating a pretrained model’s weights on task-specific data to improve performance or adapt style/behavior.
Measures how much information an observable random variable carries about unknown parameters.
A wide basin often correlated with better generalization.
Aligns transcripts with audio timestamps.
Predicting future values from past observations.
Identifying suspicious transactions.
Constraining model outputs into a schema used to call external APIs/tools safely and deterministically.
Chooses which experts process each token.
How well a model performs on new data drawn from the same (or similar) distribution as training.
Lowest possible loss.
Limiting gradient magnitude to prevent exploding gradients.
Iterative method that updates parameters in the direction of negative gradient to minimize loss.
Recovering training data from gradients.
Variability introduced by minibatch sampling during SGD.
GNN using attention to weight neighbor contributions dynamically.
Extension of convolution to graph domains using adjacency structure.
Neural networks that operate on graph-structured data by propagating information along edges.
Constraining outputs to retrieved or provided sources, often with citation, to improve factual reliability.
Rules and controls around generation (filters, validators, structured outputs) to reduce unsafe or invalid behavior.
Model-generated content that is fluent but unsupported by evidence or incorrect; mitigated by grounding and verification.
Matrix of second derivatives describing local curvature of loss.
Graphs containing multiple node or edge types with different semantics.
Probabilistic model for sequential data with latent states.
Ultra-low-latency algorithmic trading.
AI used in sensitive domains requiring compliance.
Early architecture using learned gates for skip connections.
Required human review for high-risk decisions.