Magnitude-Normalized Causal Subspace Steering
A method that projects activations onto a causal subspace and scales the intervention magnitude according to the current activation magnitude, matching the energy distribution to achieve effective control.
Timeline (last 90 days)
AROMA employs a magnitude-normalized causal subspace steering method and improves exact counting accuracy by 8.00 percentage points on procedural data it never touched.
Magnitude-normalized causal subspace steering first projects activations onto the causal subspace and then scales by the magnitude of the current activation, so that the intervention magnitude matches the actual energy distribution.
All Facts (2)
AROMA employs a magnitude-normalized causal subspace steering method and improves exact counting accuracy by 8.00 percentage points on procedural data it never touched.
50%UnverifiedMagnitude-normalized causal subspace steering first projects activations onto the causal subspace and then scales by the magnitude of the current activation, so that the intervention magnitude matches the actual energy distribution.
50%