Layer-wise Target Propagation: Efficient Component Attribution through Target Centric Propagation
cs.LG, cs.CL
Submitted: 2026-03-20
Updated: 2026-09-24
Comments: Previous title: Dual Path Attribution: Efficient Attribution for SwiGLU-Transformers through Layer-Wise Target Propagation
Code: https://github.com/lab-paper-code/dual_
License: http://creativecommons.org/licenses/by/4.0/
The gist: Understanding the internal mechanisms of transformer-based large language models (LLMs) is crucial for their reliable deployment and effective operation.
Terminology
Abstract
Understanding the internal mechanisms of transformer-based large language models (LLMs) is crucial for their reliable deployment and effective operation. While recent efforts have yielded a plethora of attribution methods attempting to balance faithfulness and computational efficiency, dense component attribution remains prohibitively expensive. In this work, we introduce Layer-wise Target Propagation (LTP), a novel framework that faithfully traces information flow on the frozen transformer in one forward and one backward pass without requiring counterfactual examples. LTP analytically decomposes and linearizes the computational structure of the Transformers into distinct pathways along which it propagates a targeted unembedding vector to receive the effective representation at each residual position. This target-centric propagation achieves O(1) time complexity with respect to the number of model components, scaling to long input sequences and dense component attribution. Extensive experiments on standard interpretability benchmarks demonstrate that LTP achieves state-of-the-art faithfulness and unprecedented efficiency compared to existing baselines.
Sources
- The Shattered Gradients Problem: If resnets are the answer, then what is the question?
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Towards Automated Circuit Discovery for Mechanistic Interpretability
- Extraction of Salient Sentences from Labelled Documents
- XAI for Transformers: Better Explanations through Conservative Propagation
- A Primer on the Inner Workings of Transformer-based Language Models
- Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Space
- The Llama 3 Herd of Models
- Mistral 7B
- DePass: Unified Feature Attributing by Simple Decomposed Forward Pass
- Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
- SmoothGrad: removing noise by adding noise
- Know What You Don't Know: Unanswerable Questions for SQuAD
- GLU Variants Improve Transformer
- RoFormer: Enhanced Transformer with Rotary Position Embedding
- Llama 2: Open Foundation and Fine-Tuned Chat Models
- Learning Important Features Through Propagating Activation Differences
- Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small
- Qwen3 Technical Report
- Qwen2.5 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