Target-Oriented Pretraining Data Selection via Neuron-Activated Graph

TL;DR

This paper introduces NAG-based Ranking, a target-oriented data selection method for language models that uses neuron impact analysis to improve pretraining effectiveness and interpretability.

Contribution

The authors propose a novel neuron impact-based framework for target data selection that outperforms existing methods and provides interpretability of the pretraining process.

Findings

NAG improves target-oriented pretraining by 4.9% on average across six benchmarks.

NAG outperforms state-of-the-art baselines by 5.3% accuracy on HellaSwag.

Deactivating NAG-selected neurons causes a 23.5% performance drop, showing their importance.

Abstract

Everyday tasks come with a target, and pretraining models around this target is what turns them into experts. In this paper, we study target-oriented language model (LM) pretraining by introducing Neuron-Activated Graph Ranking (NAG-based Ranking), a training-free and interpretable framework for target pretraining data selection. Rather than using black-box representations, our approach directly characterizes each target input by a sparse set of high-impact neurons in any off-the-shelf LLMs. Concretely, we quantify neuron impact and select the most influential neurons across layers into a compact Neuron-Activated Graph (NAG), and rank candidate data by NAG similarity to target examples. We conduct experiments across six benchmarks, where our NAG-based Ranking improves target-oriented pretraining by 4.9% on average over random sampling, and also outperforms state-of-the-art baselines by 5.3% accuracy on HellaSwag. It also remains effective under a more applicable multi-target setting, where our best setup surpasses two baselines by 1.1% and 4.1%, respectively. Furthermore, we provide a comprehensive analysis on why and how our NAG works, e.g., deactivating NAG-selected neurons (only 0.12% of all) causes a 23.5% performance collapse, and restricting NAG to the final layer incurs a 4.1% average drop, indicating that NAG captures a sparse "functional backbone" for learning target features. We release the code at https://github.com/asillycat/NAG.