CLUENet: Cluster Attention Makes Neural Networks Have Eyes

2025-12-06

AI Summary

Existing vision models—such as CNNs and Transformers—suffer from fixed receptive fields and structural complexity, limiting their ability to model irregular spatial patterns and impairing interpretability. While clustering-based methods offer semantic transparency, they exhibit low accuracy, training instability, and computational inefficiency. To address these limitations, we propose CLUENet: a highly transparent vision network grounded in clustering-based attention. Its key innovations include global soft–hard hybrid aggregation, temperature-scaled cosine attention, gated residual connections, inter-block hard-shared feature distribution, and differentiable clustering pooling. This design jointly preserves local modeling fidelity and semantic flexibility while ensuring efficient gradient propagation and architectural traceability. Evaluated on CIFAR-100 and Mini-ImageNet, CLUENet significantly outperforms state-of-the-art clustering models and mainstream vision architectures, achieving balanced advances in accuracy, inference efficiency, and interpretability.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Clustering
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Web Mining and Content Analysis: Normalization, clustering, classification, and summarization of Web text
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Despite the success of convolution- and attention-based models in vision tasks, their rigid receptive fields and complex architectures limit their ability to model irregular spatial patterns and hinder interpretability, therefore posing challenges for tasks requiring high model transparency. Clustering paradigms offer promising interpretability and flexible semantic modeling, but suffer from limited accuracy, low efficiency, and gradient vanishing during training. To address these issues, we propose CLUster attEntion Network (CLUENet), an transparent deep architecture for visual semantic understanding. We propose three key innovations include (i) a Global Soft Aggregation and Hard Assignment with a Temperature-Scaled Cosin Attention and gated residual connections for enhanced local modeling, (ii) inter-block Hard and Shared Feature Dispatching, and (iii) an improved cluster pooling strategy. These enhancements significantly improve both classification performance and visual interpretability. Experiments on CIFAR-100 and Mini-ImageNet demonstrate that CLUENet outperforms existing clustering methods and mainstream visual models, offering a compelling balance of accuracy, efficiency, and transparency.

Problem

• Enhances interpretability and flexible semantic modeling in vision tasks
• Addresses limited accuracy and efficiency in clustering-based neural networks
• Improves classification performance and transparency for visual understanding

Innovation

• Global soft aggregation with temperature-scaled attention
• Inter-block hard and shared feature dispatching
• Improved cluster pooling strategy for interpretability