framework
active
framework:interference-minimization-view-of-trainingInterference minimization view of training
Average-case vs worst-case interference objective conceptualization.
Neighborhood — ranked by edge-count
Papers (1)
paper
- Adversarial Examples Are Not Bugs, They Are Superpositionintroducesmentions
Concepts (1)
concept
- Adversarial traininganalogous_toIntervention used to control robustness.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- The progressive reduction of error (stress) as cells move toward their target positions.
- The drive to reduce expected ambiguity about outcomes given states, leading to seeking well-lit, informative environments.
- The principle that agents must minimise prediction error (surprisal) to persist.
- Frequent-task updates overwrite rare-task features.
- Method that optimizes activation interventions so that resulting behaviors trace M_y, recovering activation paths that follow M_h curvature.
- Core principle: acting to maximize value is equivalent to minimizing surprise by sampling environment to conform to expectations.
- Ethical implication about the nature of AI training experience if the thesis holds
- When non-orthogonal features cause logistic regression to identify a suboptimal probe direction