Results for "conditional computation"

Physical form contributes to computation.

Probabilistic graphical model for structured prediction.

Routes inputs to subsets of parameters for scalable capacity.

Enables external computation or lookup.

Reduction in uncertainty achieved by observing a variable; used in decision trees and active learning.

Attention where queries/keys/values come from the same sequence, enabling token-to-token interactions.

Attacks that infer whether specific records were in training data, or reconstruct sensitive training examples.

Methods to protect model/data during inference (e.g., trusted execution environments) from operators/attackers.

Techniques to handle longer documents without quadratic cost.

Fundamental recursive relationship defining optimal value functions.

Formal framework for sequential decision-making under uncertainty.

Graphical model expressing factorization of a probability distribution.

Matrix of first-order derivatives for vector-valued functions.

Eliminating variables by integrating over them.

Computing end-effector position from joint angles.

The relationship between inputs and outputs changes over time, requiring monitoring and model updates.

When information from evaluation data improperly influences training, inflating reported performance.

Training objective where the model predicts the next token given previous tokens (causal modeling).

Predicts masked tokens in a sequence, enabling bidirectional context; often used for embeddings rather than generation.

Quantifies shared information between random variables.

Models that learn to generate samples resembling training data.

Directed acyclic graph encoding causal relationships.

Assigning a role or identity to the model.

Prompt augmented with retrieved documents.

Model relies on irrelevant signals.

Inferring human goals from behavior.

Differences between training and deployed patient populations.

Trend reversal when data is aggregated improperly.