PruneFuse: Efficient Data Selection via Weight Pruning and Network Fusion

arXiv cs.LG / 3/30/2026

📰 NewsIdeas & Deep AnalysisModels & Research

共有:

Key Points

PruneFuse is a two-stage deep-learning strategy that improves training efficiency by using structured pruning to create a smaller network for informative data selection.
After the pruned model is trained to identify the most informative samples, its learned knowledge is fused back into the original network to speed and strengthen overall learning.
The method is designed to reduce the high computational cost and scalability limitations of traditional data selection approaches that often require substantial annotation effort.
Experiments across multiple datasets reportedly show PruneFuse lowers data-selection compute, improves performance versus baselines, and accelerates end-to-end training.

Abstract

Efficient data selection is crucial for enhancing the training efficiency of deep neural networks and minimizing annotation requirements. Traditional methods often face high computational costs, limiting their scalability and practical use. We introduce PruneFuse, a novel strategy that leverages pruned networks for data selection and later fuses them with the original network to optimize training. PruneFuse operates in two stages: First, it applies structured pruning to create a smaller pruned network that, due to its structural coherence with the original network, is well-suited for the data selection task. This small network is then trained and selects the most informative samples from the dataset. Second, the trained pruned network is seamlessly fused with the original network. This integration leverages the insights gained during the training of the pruned network to facilitate the learning process of the fused network while leaving room for the network to discover more robust solutions. Extensive experimentation on various datasets demonstrates that PruneFuse significantly reduces computational costs for data selection, achieves better performance than baselines, and accelerates the overall training process.