Out-of-Distribution (OOD) Generalization

Machine learning generalization when training and test distributions differ; linked to causal invariance.

Neighborhood — ranked by edge-count

paper

claim

EI and normalized EI could serve as a unified metric for out-of-distribution generalization.
supports
Conjecture that maximizing EI yields causal representations invariant to distribution shifts.

concept

Out-of-Distribution Probe Generalization
related_to
The capacity of a probe trained on one true/false dataset to accurately classify statements from topically and structurally different datasets

hypothesis

cosine ≥ 0.65 · no typed edge

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

Out-of-distribution Steeringconcept0.796
Steering toward directions that deviate from the model's typical generation distribution, causing rapid coherency collapse
Generalizationconcept0.789
Ability to apply learned solutions to novel circumstances.
Weird Generalizationconcept0.782
Phenomenon where fine-tuning shifts behavior on tasks unrelated to training data, framed through personas
CoT improves in-distribution but may harm out-of-distribution generalizationclaim0.781
Interpretation of scope generalization results
Scope generalization: CoT boosts 2-digit in-distribution but worsens 3-4 digit OODfinding0.773
CoT increases dr for OOD operands.
Generalisationconcept0.773
Ability to respond appropriately to novel situations based on past regularities; fundamental to learning and intelligence.
generalization (abstraction)concept0.770
Abstracting from specific memories (e.g., specific leaves) to general lessons (food).
In-distribution re-alignment more effectively reverses the original fine-tuning procedure, while out-of-distribution re-alignment mainly suppresses misalignment generalizationclaim0.754
Code-realigned model writes less insecure code than health-realigned model after identical steps