Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Association and Consolidation: Evolutionary Memory-Enhanced Incremental Multi-View Clustering

Zisen Kong, Bo Zhong, Pengyuan Li, Dongxia Chang, Yiming Wang, Yongyong Chen

TL;DR

EMIMC addresses the stability-plasticity dilemma in incremental multi-view clustering by emulating hippocampus-prefrontal cortex memory dynamics. It introduces three modules—Rapid Associative Module (RAM) for fast integration of new views, Cognitive Forgetting Module (CFM) for adaptive forgetting, and Knowledge Consolidation Memory (KCM) for stabilizing knowledge via a two-slice temporal tensor and a low-rank ARMR constraint. The overall objective combines $\mathcal{L}_{recon}$, $\alpha\mathcal{L}_{associate}$, and $\beta\mathcal{L}_{consolidate}$, optimized with ADMM across initial and incremental views. Empirical results on six datasets show substantial accuracy gains over baselines, robustness to view order, and scalable performance with growing views. This approach offers a practical framework for evolving multi-view data where new modalities arrive over time.

Abstract

Incremental multi-view clustering aims to achieve stable clustering results while addressing the stability-plasticity dilemma (SPD) in view-incremental scenarios. The core challenge is that the model must have enough plasticity to quickly adapt to new data, while maintaining sufficient stability to consolidate long-term knowledge. To address this challenge, we propose a novel Evolutionary Memory-Enhanced Incremental Multi-View Clustering (EMIMC), inspired by the memory regulation mechanisms of the human brain. Specifically, we design a rapid association module to establish connections between new and historical views, thereby ensuring the plasticity required for learning new knowledge. Second, a cognitive forgetting module with a decay mechanism is introduced. By dynamically adjusting the contribution of the historical view to optimize knowledge integration. Finally, we propose a knowledge consolidation module to progressively refine short-term knowledge into stable long-term memory using temporal tensors, thereby ensuring model stability. By integrating these modules, EMIMC achieves strong knowledge retention capabilities in scenarios with growing views. Extensive experiments demonstrate that EMIMC exhibits remarkable advantages over existing state-of-the-art methods.

Association and Consolidation: Evolutionary Memory-Enhanced Incremental Multi-View Clustering

TL;DR

EMIMC addresses the stability-plasticity dilemma in incremental multi-view clustering by emulating hippocampus-prefrontal cortex memory dynamics. It introduces three modules—Rapid Associative Module (RAM) for fast integration of new views, Cognitive Forgetting Module (CFM) for adaptive forgetting, and Knowledge Consolidation Memory (KCM) for stabilizing knowledge via a two-slice temporal tensor and a low-rank ARMR constraint. The overall objective combines , , and , optimized with ADMM across initial and incremental views. Empirical results on six datasets show substantial accuracy gains over baselines, robustness to view order, and scalable performance with growing views. This approach offers a practical framework for evolving multi-view data where new modalities arrive over time.

Abstract

Incremental multi-view clustering aims to achieve stable clustering results while addressing the stability-plasticity dilemma (SPD) in view-incremental scenarios. The core challenge is that the model must have enough plasticity to quickly adapt to new data, while maintaining sufficient stability to consolidate long-term knowledge. To address this challenge, we propose a novel Evolutionary Memory-Enhanced Incremental Multi-View Clustering (EMIMC), inspired by the memory regulation mechanisms of the human brain. Specifically, we design a rapid association module to establish connections between new and historical views, thereby ensuring the plasticity required for learning new knowledge. Second, a cognitive forgetting module with a decay mechanism is introduced. By dynamically adjusting the contribution of the historical view to optimize knowledge integration. Finally, we propose a knowledge consolidation module to progressively refine short-term knowledge into stable long-term memory using temporal tensors, thereby ensuring model stability. By integrating these modules, EMIMC achieves strong knowledge retention capabilities in scenarios with growing views. Extensive experiments demonstrate that EMIMC exhibits remarkable advantages over existing state-of-the-art methods.

Paper Structure

This paper contains 22 sections, 1 theorem, 28 equations, 11 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries
  3. Incremental multi-view clustering
  4. Tensor for knowledge consolidation
  5. The Proposed method
  6. Overview and Motivation
  7. Plasticity: New Knowledge Acquisition
  8. Stability: Knowledge Consolidation and Evolution
  9. The Overall Objective Function
  10. Model optimization
  11. Initial view processing (t = 1).
  12. Updates to subsequent views (t > 1).
  13. Complexity Analysis
  14. Experiments
  15. Experimental Results and Analysis
  16. ...and 7 more sections

Key Result

Lemma 1

Let tensor $\mathcal{M}\in\mathbb{R}^{n_1\times n_2 \times n_3}$ with t-SVD. The non-concave tensor rank minimization regularization is solved by: The optimal solution of Eq.(tensor_Z2) can be written as where ifft denotes the Inverse Fast Fourier Transform. The function $f(\mathcal{S}_f^k)$ is a f-diagonal tensor, which is

Figures (11)

  • Figure 1: The framework of EMIMC. The model achieves the retention of historical knowledge and the learning of new knowledge through three core modules: Rapid Associative, Cognitive Forgetting, and Knowledge Consolidation Module.
  • Figure 2: IWTSN (ACC = 73.31%)
  • Figure 3: Ours (ACC = 92.28%)
  • Figure 5: Flower17
  • Figure 6: GRAZ02
  • ...and 6 more figures

Theorems & Definitions (2)

  • Definition 1
  • Lemma 1