Relative performance improvement (Δp)

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

• Conditional Mean Score Improvement (metric)concept0.754
  Score delta between last and first attempt for multi-attempt responses, measuring correction effectiveness
• Representational dynamics aligned with reward improvement in most RL tasks.finding0.733
  Secondary empirical result: CE-based representational changes correlate with task success.
• Reward improvementconcept0.708
  The increase in reward during training, whose dynamics align with those of causal emergence in successful agents.
• Representational dynamics of causal emergence align with reward improvement in most tasks.finding0.702
  The trajectory of causal emergence through training mirrors the reward improvement curve across the majority of tested environments.
• Strength comparison pair (3,7) with |Δα|=4 outperforms pair (3,5) with |Δα|=2, indicating graded sensitivity to perturbation magnitudefinding0.702
  Shows that introspective accuracy scales with injection strength difference, not binary detection
• GRPO (Group Relative Policy Optimization)method0.698
  RL algorithm used to train the activation verbalizer on open models; samples group of candidate descriptions and applies policy optimization.
• The direction of information increase is relative to the observer or user of the computationclaim0.697
  Example: 3×5→15 is a natural computation, but 15→3×5 (prime factorization) is also useful, showing that the 'gain' depends on the choice of normal form.
• Using only loss-scale balancing (log transformation) yields Δp = +0.06±0.09 on NYUv2.finding0.696
  Ablation study component effectiveness.