Enum AttentionMatchingMode

Namespace

AiDotNet.KnowledgeDistillation.Strategies

Assembly

AiDotNet.dll

Defines how to match attention patterns between teacher and student.

public enum AttentionMatchingMode

Fields

Cosine = 2

Cosine similarity - focuses on direction/pattern rather than magnitude.

KL = 1

KL Divergence - treats attention as probability distribution, preserves structure.

MSE = 0

Mean Squared Error - simple, fast, treats all attention weights equally.