Hierarchical Feature-level Reverse Propagation for Post-Training Neural Networks

2025-06-08

AI Summary

To address the weak interpretability, safety assurance challenges, and high training coupling inherent in end-to-end autonomous driving models, this paper proposes a hierarchical decoupling post-training optimization framework. It reconstructs intermediate feature maps by back-propagating from output labels—formulating feature inversion as a well-posed linear system (either a linear equation system or least-squares problem), thereby achieving gradient-free, feature-level inverse supervision for the first time. This enables generation of surrogate supervision signals that support independent, transparent, and interpretable optimization of individual network modules. The method overcomes the parameter coupling limitations of conventional backpropagation. Evaluated on multiple image classification benchmarks, it significantly improves generalization performance and inference efficiency, while maintaining theoretical rigor and engineering practicality.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search
• Machine Learning: Optimization

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

End-to-end autonomous driving has emerged as a dominant paradigm, yet its highly entangled black-box models pose significant challenges in terms of interpretability and safety assurance. To improve model transparency and training flexibility, this paper proposes a hierarchical and decoupled post-training framework tailored for pretrained neural networks. By reconstructing intermediate feature maps from ground-truth labels, surrogate supervisory signals are introduced at transitional layers to enable independent training of specific components, thereby avoiding the complexity and coupling of conventional end-to-end backpropagation and providing interpretable insights into networks' internal mechanisms. To the best of our knowledge, this is the first method to formalize feature-level reverse computation as well-posed optimization problems, which we rigorously reformulate as systems of linear equations or least squares problems. This establishes a novel and efficient training paradigm that extends gradient backpropagation to feature backpropagation. Extensive experiments on multiple standard image classification benchmarks demonstrate that the proposed method achieves superior generalization performance and computational efficiency compared to traditional training approaches, validating its effectiveness and potential.

Problem

• Improving interpretability and safety in autonomous driving models
• Enabling independent training of neural network components
• Formalizing feature-level reverse computation as optimization problems

Innovation

• Hierarchical feature-level reverse propagation
• Surrogate supervisory signals for training
• Feature backpropagation as optimization problems