Term #2,252

contextual weights

Origin matter: BRT-PAT-2-PROV

Claim Term contextual weights
Reference Case 1 BRT_PAT-2-PROV
Date Added 7/14/25
Created 7/14/25, 9:13 PM
Modified 7/14/25, 9:13 PM
Full Desc
“Contextual weights” refers to numerical values computed by an attention mechanism or similar model component that represent the relative importance or relevance of elements within an input sequence, based on their relationship to one another in a given context. Contextual weights are typically derived from comparisons between query vectors and key vectors, or the like, and are used to scale corresponding value vectors or features during model inference. These weights allow a model to emphasize or de-emphasize particular elements of the input when generating intermediate representations or outputs, such as in language modeling, classification, summarization, or the like. Contextual weights may vary dynamically across tasks, sequences, or inference steps, and may be computed using dot product attention, additive attention, or the like. (Defined in conjunction with ChatGPT 4o Version, July 14, 2025.)

Applications using this term

Matter Usage Notes Actions
BRT-PAT-2-PROV Defined App terms