finding

active

finding:interventions-along-activation-manifold-m-h-yield-behavioral-trajectories-following-behavior-manifold-m-y-and-vice-versa-bidirectional-relationship-demonstrated-across-language-models-and-video-world-models

Interventions along activation manifold M_h yield behavioral trajectories following behavior manifold M_y, and vice versa — bidirectional relationship demonstrated across language models and video world models.

Central empirical result showing causal coupling between representation and behavior geometry across multiple substrates and modalities.

Source paper

extracted_from

Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior

(2026) · Daniel Wurgaft · Can Rager · Matthew Kowal · Vasudev Shyam +12

Neighborhood — ranked by edge-count

Claims (3)

claim

Geometry in neural representation is not merely incidental, but is in fact the proper object for enabling principled control via intervention on internals.
supports
Core interpretive assertion: geometric structure is causally load-bearing, not epiphenomenal.
Linear steering cuts through off-manifold regions and hence produces unnatural outputs.
supports
Attribution of failure to Euclidean assumption.
There is a bidirectional relationship between the geometry of representation and behavior across tasks and modalities.
supports
Author’s interpretive claim that the shared geometry is general and robust.

Hypotheses (1)

hypothesis

Neural representation geometry causally shapes behavior; interventions respecting that geometry will yield natural trajectories.
supports
Central hypothesis tested via manifold steering experiments across language models and video world models.

Communities (3)

community

Manifold-aware concept steering in neural representations
members_of
Explores geometry of activation/behavior manifolds to enable selective, non-destructive concept interventions.
Geometric concept representations in neural networks
members_of
Concepts encoded as curved manifolds and circular structures in LLM activation spaces.
Manifold isometry between representations and behavior
members_of
Neural network activation and behavior manifolds maintain geometric correspondence, enabling intervention optimization across language models and vision tasks.

Methods (1)

method

linear steering
contradicts
Typical approach that adds a scaled steering vector to representations; the paper argues this is mismatched with actual representation geometry.

Concepts (1)

concept

Euclidean geometry assumption in steering
contradicts
Linear steering implicitly assumes a flat, Euclidean activation space, leading to off-manifold excursions.

Questions (2)

question

does that structure causally shape behavior?
answered_by
Opening question: does the rich geometric structure of neural representations have a causal role in behavior?
Does the geometric structure of activation space causally shape neural network behavior?
answered_by
Central research question driving the work.

Related by similarity (8)

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.

Optimizing interventions in activation space to produce paths along M_y recovers activation trajectories that trace the curvature of M_h.finding0.853
Demonstrates bidirectional causal link: behavior manifold geometry can be recovered by optimizing in representation space.
We hypothesize that interventions that respect the geometry of activation space will yield behaviors close to those the model exhibits naturallyhypothesis0.838
The core testable hypothesis driving the experimental design
optimization of interventions to follow behavior manifold M_ymethod0.831
Method that optimizes activation interventions so that resulting behaviors trace M_y, recovering activation paths that follow M_h curvature.
Activation manifolds and behavior manifolds are approximately isometric across cyclic and sequential concepts.claim0.822
Analogous alignment between representation manifold and behavior manifold is found across months, letters, ages, and synthetic in-context learning tasks in language models.finding0.815
Generalization finding from the full paper extending beyond days-of-week to other structured concepts.
Representation geometry causally shapes behavior; activation and behavior manifolds are approximately isometric.claim0.810
There is a clear bidirectional relationship between the geometry of behavior and representation: steering along representation manifolds follows behavior manifolds, and vice versa.claim0.807
The paper's finding that the alignment holds in both directions — from representation to behavior and from behavior back to representation space.
Linear interventions across voids in activation space produce incoherent output, while following the manifold curve produces smooth control.claim0.802
General principle derived from the Mountain Car experiment: curved manifold-following yields coherent manipulation, linear shortcuts fail.