Beyond Linear Concepts: Discovering and Aligning Non-Linear Concept Manifolds in Large Language Models
cs.LG, cs.CL
Submitted: 2026-10-01
Updated: 2026-10-01
Comments: 24 pages, 13 figures. Code: https://anonymous.4open.science/r/NLMCD-NLP-C5E7
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Not All Language Model Features Are One-Dimensionally Linear
- Scaling and evaluating sparse autoencoders
- The Llama 3 Herd of Models
- The Platonic Representation Hypothesis
- Tulu 3: Pushing Frontiers in Open Language Model Post-Training
- Language Models Encode Numbers Using Digit Representations in Base 10
- UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction
- BERT Rediscovers the Classical NLP Pipeline
- Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior
- The Shape of Beliefs: Geometry, Dynamics, and Interventions along Representation Manifolds of Language Models' Posteriors
- Qwen3 Technical Report
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks