ConceptSparse Feature Decomposition
Sparse Feature Decomposition
A technique that uses sparse autoencoders to extract interpretable sparse feature directions from high-dimensional neural network activations, helping distinguish features that carry critical information from noise.
Timeline (last 90 days)
Oct 16
The core idea of the TITAN framework is to shift the optimization of translation capability from full-parameter fine-tuning to causal circuit localization and targeted optimization, integrating three technical paths: causal intervention, sparse feature decomposition, and targeted preference optimization.
Unverified50%