Efficient Hybrid Network: Inducting Scattering Features

TL;DR

The paper introduces E-HybridNet, a scattering-based hybrid network that outperforms traditional models across datasets by integrating scattering features with learned filters through a novel architecture, balancing flexibility and stability.

Contribution

It presents the first scattering-based hybrid network that consistently surpasses conventional models, using Hybrid Fusion Blocks to embed scattering features effectively.

Findings

Outperforms conventional networks on multiple datasets

Retains good generalization in data-limited scenarios

Combines flexibility of learned features with stability of scattering representations

Abstract

Recent work showed that hybrid networks, which combine predefined and learnt filters within a single architecture, are more amenable to theoretical analysis and less prone to overfitting in data-limited scenarios. However, their performance has yet to prove competitive against the conventional counterparts when sufficient amounts of training data are available. In an attempt to address this core limitation of current hybrid networks, we introduce an Efficient Hybrid Network (E-HybridNet). We show that it is the first scattering based approach that consistently outperforms its conventional counterparts on a diverse range of datasets. It is achieved with a novel inductive architecture that embeds scattering features into the network flow using Hybrid Fusion Blocks. We also demonstrate that the proposed design inherits the key property of prior hybrid networks -- an effective generalisation in data-limited scenarios. Our approach successfully combines the best of the two worlds: flexibility and power of learnt features and stability and predictability of scattering representations.