Inference and learning in sparse autoencoders as natural gradient flow

arXiv:2610.07389 · cs.LG, cs.AI · Submitted 2026-10-05 · Read on arXiv

cs.LG, cs.AI

Submitted: 2026-10-05

Updated: 2026-10-05

Code: https://github.com/hadivafaii/BeFOND

Terminology

• Gradient Flow — 1× in this paper · explained in 2 episodes

Sources

• Information-Geometric Signatures of Nonconservative Driving
• Stop Probing, Start Coding: Why Linear Probes and Sparse Autoencoders Fail at Compositional Generalisation
• BatchTopK Sparse Autoencoders
• SynthSAEBench: Evaluating Sparse Autoencoders on Scalable Realistic Synthetic Data
• Gemma 2: Improving Open Language Models at a Practical Size
• Auto-Encoding Variational Bayes
• k-Sparse Autoencoders
• The information geometry of large language models is shared, learned, and controllable
• Jumping Ahead: Improving Reconstruction Fidelity with JumpReLU Sparse Autoencoders

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