FMCE-Net++: Feature Map Convergence Evaluation and Training

TL;DR

FMCE-Net++ introduces a training framework that leverages feature map convergence evaluation to improve neural network performance without changing architecture or data, validated across multiple datasets.

Contribution

It proposes FMCE-Net++, a novel training method integrating a pretrained FMCE module with a dynamic loss to enhance model accuracy.

Findings

Achieves accuracy improvements of +1.16 percentage points on CIFAR-10 with ResNet-50.

Attains +1.08 percentage points accuracy gain on CIFAR-100 with ShuffleNet v2.

Demonstrates consistent performance boosts across diverse datasets and architectures.

Abstract

Deep Neural Networks (DNNs) face interpretability challenges due to their opaque internal representations. While Feature Map Convergence Evaluation (FMCE) quantifies module-level convergence via Feature Map Convergence Scores (FMCS), it lacks experimental validation and closed-loop integration. To address this limitation, we propose FMCE-Net++, a novel training framework that integrates a pretrained, frozen FMCE-Net as an auxiliary head. This module generates FMCS predictions, which, combined with task labels, jointly supervise backbone optimization through a Representation Auxiliary Loss. The RAL dynamically balances the primary classification loss and feature convergence optimization via a tunable \Representation Abstraction Factor. Extensive experiments conducted on MNIST, CIFAR-10, FashionMNIST, and CIFAR-100 demonstrate that FMCE-Net++ consistently enhances model performance without architectural modifications or additional data. Key experimental outcomes include accuracy gains of $+1.16$ pp (ResNet-50/CIFAR-10) and $+1.08$ pp (ShuffleNet v2/CIFAR-100), validating that FMCE-Net++ can effectively elevate state-of-the-art performance ceilings.