Symbolic Mechanistic Data Attribution: Tracing Training Influence to Learned Behavioral Policies

arXiv:2606.29171 · cs.LG, cs.AI, cs.CL · Submitted 2026-06-28 · Read on arXiv

cs.LG, cs.AI, cs.CL

Submitted: 2026-06-28

Updated: 2026-10-15

Code: https://github.com/jbloomAus/SAELens

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Symbolic Mechanistic Data Attribution — 2× in this paper
• Fine-Tuning — 1× in this paper · explained in 5 episodes
• Influence Functions — 1× in this paper · explained in 2 episodes
• Interpretability — 1× in this paper · explained in 15 episodes
• Mechanistic Interpretability — 1× in this paper · explained in 3 episodes
• Sparse Autoencoder (SAE) — 1× in this paper · explained in 2 episodes

Sources

• Mechanistic Data Attribution: Tracing the Training Origins of Interpretable LLM Units
• Sparse Autoencoders Find Highly Interpretable Features in Language Models
• Studying Large Language Model Generalization with Influence Functions
• Beyond Accuracy: Introducing a Symbolic-Mechanistic Approach to Interpretable Evaluation
• Concept Influence: Leveraging Interpretability to Improve Performance and Efficiency in Training Data Attribution
• In-context Learning and Induction Heads
• Circuit Distillation
• Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs

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