Feature Imitating Networks

TL;DR

The paper introduces Feature-Imitating-Networks (FINs), neural networks initialized to approximate statistical features, which achieve superior performance in signal processing tasks with less data and fine-tuning.

Contribution

It presents a novel neural network initialization method that encodes statistical features, improving downstream task performance and bridging domain expertise with machine learning.

Findings

FINs outperform traditional networks in signal processing tasks.

FIN ensembles require less data and fine-tuning.

FINs effectively incorporate domain-specific features.

Abstract

In this paper, we introduce a novel approach to neural learning: the Feature-Imitating-Network (FIN). A FIN is a neural network with weights that are initialized to reliably approximate one or more closed-form statistical features, such as Shannon's entropy. In this paper, we demonstrate that FINs (and FIN ensembles) provide best-in-class performance for a variety of downstream signal processing and inference tasks, while using less data and requiring less fine-tuning compared to other networks of similar (or even greater) representational power. We conclude that FINs can help bridge the gap between domain experts and machine learning practitioners by enabling researchers to harness insights from feature-engineering to enhance the performance of contemporary representation learning approaches.