ILLC: Iterative Layer-by-Layer Compression for Enhancing Structural Faithfulness in SpArX

2025-03-05

AI Summary

To address cumulative information loss caused by global simplification in multi-layer deep neural network compression, this paper proposes a layer-wise iterative compression method tailored for argumentative eXplainable AI (XAI). Grounded in the SpArX argumentation graph modeling framework, the method integrates inter-layer error propagation analysis, layer-specific sensitivity-weighted compression, and residual feedback correction to enable dynamic cross-layer error compensation. Unlike conventional one-shot compression approaches, our method preserves model compactness while significantly enhancing dual fidelity—both input-output mapping fidelity and internal argumentative structure fidelity. Experimental evaluation on a breast cancer diagnosis dataset demonstrates a 32.7% reduction in input-output infidelity and a 28.4% decrease in structural infidelity, alongside markedly improved consistency in attack-support relationships among arguments.

Technology Category

• Knowledge Representation and Reasoning: Argumentation
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Machine Learning: Learning on the Edge & Model Compression

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

In the field of Explainable Artificial Intelligence (XAI), argumentative XAI approaches have been proposed to represent the internal reasoning process of deep neural networks in a more transparent way by interpreting hidden nodes as arguements. However, as the number of layers increases, existing compression methods simplify all layers at once, which lead to high accumulative information loss. To compensate for this, we propose an iterative layer-by-layer compression technique in which each layer is compressed separately and the reduction error in the next layer is immediately compensated for, thereby improving the overall input-output and structural fidelity of the model. Experiments on the Breast Cancer Diagnosis dataset show that, compared to traditional compression, the method reduces input-output and structural unfaithfulness, and maintains a more consistent attack-support relationship in the Argumentative Explanation scheme. This is significant because it provides a new way to make complex MLP models more compact while still conveying their internal inference logic without distortion.

Problem

• Reduces information loss in deep neural network compression
• Improves structural and input-output fidelity in XAI
• Maintains consistent attack-support relationships in explanations

Innovation

• Iterative layer-by-layer compression technique
• Separate compression with error compensation
• Enhanced input-output and structural fidelity