Multi-View Incremental Learning with Structured Hebbian Plasticity for Enhanced Fusion Efficiency

TL;DR

This work tackles the challenge of multi-view learning under sequentially arriving views by introducing MVIL, a biologically inspired framework featuring structured Hebbian plasticity and synaptic partition learning. MVIL uses streaming view representation learning with a shared two-layer GCN to fuse new views with past representations, while Hebbian reinforcement strengthens consistent cross-view connections and synaptic partitioning reduces drastic weight changes. Empirical results on six multi-view graph datasets show MVIL achieving state-of-the-art performance in semi-supervised node classification, with favorable time and space efficiency and robust knowledge accumulation across streaming views. The approach highlights the potential of brain-inspired plasticity for robust continual and incremental learning in dynamic, real-world multi-view data scenarios.

Abstract

The rapid evolution of multimedia technology has revolutionized human perception, paving the way for multi-view learning. However, traditional multi-view learning approaches are tailored for scenarios with fixed data views, falling short of emulating the intricate cognitive procedures of the human brain processing signals sequentially. Our cerebral architecture seamlessly integrates sequential data through intricate feed-forward and feedback mechanisms. In stark contrast, traditional methods struggle to generalize effectively when confronted with data spanning diverse domains, highlighting the need for innovative strategies that can mimic the brain's adaptability and dynamic integration capabilities. In this paper, we propose a bio-neurologically inspired multi-view incremental framework named MVIL aimed at emulating the brain's fine-grained fusion of sequentially arriving views. MVIL lies two fundamental modules: structured Hebbian plasticity and synaptic partition learning. The structured Hebbian plasticity reshapes the structure of weights to express the high correlation between view representations, facilitating a fine-grained fusion of view representations. Moreover, synaptic partition learning is efficient in alleviating drastic changes in weights and also retaining old knowledge by inhibiting partial synapses. These modules bionically play a central role in reinforcing crucial associations between newly acquired information and existing knowledge repositories, thereby enhancing the network's capacity for generalization. Experimental results on six benchmark datasets show MVIL's effectiveness over state-of-the-art methods.

Figures (8)

• Figure 1: Sensory signals reach the brain sequentially, and in parallel with this, the view data increases dynamically.
• Figure 2: An overview of the proposed framework MVIL, which is drawn by Figdraw. MVIL introduces a weight adjustment strategy that incorporates structured Hebbian plasticity and synaptic partition learning, aiming to allow the model to learn with higher biological fidelity. The structured Hebbian plasticity reshapes the weight structure by reinforcing the consistent information between the old and new views; while the latter slows down weight changes by suppressing some of the synapses.
• Figure 3: A comparison of MVIL streaming input and single-view learning performance.
• Figure 4: The time-space cost of all compared algorithms.
• Figure 5: The initial features and learned fused representations for each view of the NGs.