Published on:
4 December 2023
Primary Category:
Machine Learning
Paper Authors:
Jungwon Choi,
Seongho Keum,
EungGu Yun,
Byung-Hoon Kim,
Juho Lee
Proposes ST-MAE, a self-supervised model for dynamic functional connectivity graphs
Pre-trains on 40K+ UK Biobank fMRI samples to learn useful representations
Outperforms baselines in downstream tasks like classification and regression
Particularly effective for psychiatric diagnosis with limited labeled data
Validated extensively on diverse fMRI datasets like ABCD, HCP, ABIDE
Self-Supervised Learning for Functional Connectivity Networks
This paper proposes a self-supervised learning framework tailored for functional connectivity networks from fMRI data. It introduces Spatio-Temporal Masked Autoencoder (ST-MAE) to effectively capture both spatial graph structure and temporal dynamics. Pre-trained on a large UK Biobank fMRI dataset, it demonstrates superior performance over baselines in downstream tasks including gender/age prediction and psychiatric diagnosis.
Masked modeling for understanding motion in crowded scenes
Person re-identification with masked autoencoders
Self-supervised video representation learning using masked reconstruction
Diffusion Models as Masked Autoencoders: A New Approach for Learning Visual Representations
Learning representations of biosignals with masked autoencoders
Self-supervised learning for segmenting elevation data
No comments yet, be the first to start the conversation...
Sign up to comment on this paper