Neuron-Level Architecture Growth: A Controlled Evaluation for EEG Time-Series Decoding

2026-09-27

AI Summary

This study evaluates the practical gains of neuron-level architecture growth over fixed-width models for EEG-based motor imagery decoding. Methodologically, leveraging twelve datasets and three convolutional backbones, we propose a neuron selection mechanism utilizing dynamic thresholds derived from singular value decomposition, while employing quantified skip rates to guide the design of lightweight compact models. Our results demonstrate that effective ranking criteria are crucial for architecture growth. Notably, ShallowFBCSPNet achieves a 2.9 percentage point accuracy improvement with only half the original parameter count. These findings validate the significant advantages of the proposed approach in realizing efficient electroencephalogram decoding.

Abstract

Convolutional EEG decoders are trained at a fixed width, usually set by their authors on other data. Growing methods add neurons during training where the loss could decrease the most, but whether they improve compared to a reference width is untested on EEG. Here, we grow three convolutional backbones on 12 motor-imagery datasets under three protocols and compare each with its reference model per subject. The growing ShallowFBCSPNet scores 2.9 points above its reference model with only half the parameters (0.57x), SCCNet changes by at most 1.2 points. Deep4Net growing models show decreased accuracy, but they require adaptation that prevent to compare faithfully the results. These differences follow the selection step, which keeps a candidate neuron relying on a dynamic threshold from singular values decomposition. Overall, these results suggest that growth helps when its criterion can rank the candidate neurons, and that the rate of skipped neuron addition tells where a decoder can be grown small from scratch.

Problem

• EEG decoding
• architecture growth
• convolutional neural networks
• neuron-level
• motor imagery

Innovation

• Neuron-Level Architecture Growth
• EEG Decoding
• Singular Value Decomposition
• Dynamic Threshold
• Motor Imagery