Local Activation Norm Rescaling

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• L0 Norm of Feature Activationsconcept0.794
  Average number of nonzero feature entries per input; primary measure of activation sparsity in the autoencoder
• Maximum gradient norm scalingconcept0.729
  Scaling aggregated gradient by the maximum gradient norm among tasks.
• Activation Similarityconcept0.718
  Model-independent feature comparison based on correlating activation vectors across a fixed diverse dataset
• Other-Referencing Activationsconcept0.716
  Latent model activations when processing inputs framed from another agent's perspective
• Activation Cappingmethod0.716
  Clamping activations along the Assistant Axis to remain above a minimum threshold (25th percentile), introduced as a stabilization method
• Fine-Tuning Threshold Recalibrationmethod0.715
  Re-running probabilistic bisection on each fine-tuned checkpoint to normalize first-attempt difficulty
• Adaptive Beta Softmax Scalingmethod0.708
  Implementation detail weighting softmax by log(n_memories) to prevent down-weighting of attention values as memory set grows.
• Threshold-like activation assumptionconcept0.708
  Assumption that small anchor changes can produce sharp performance shifts when conditions are favorable.