Saved in:
Bibliographic Details
Main Authors: Xie, Yuhan, Cappelletti, William, Shoaran, Mahsa, Frossard, Pascal
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.10147
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911105333329920
author Xie, Yuhan
Cappelletti, William
Shoaran, Mahsa
Frossard, Pascal
author_facet Xie, Yuhan
Cappelletti, William
Shoaran, Mahsa
Frossard, Pascal
contents Deep neural networks for time series must capture complex temporal patterns, to effectively represent dynamic data. Self- and semi-supervised learning methods show promising results in pre-training large models, which -- when finetuned for classification -- often outperform their counterparts trained from scratch. Still, the choice of pretext training tasks is often heuristic and their transferability to downstream classification is not granted, thus we propose a novel semi-supervised pre-training strategy to enforce latent representations that satisfy the Neural Collapse phenomenon observed in optimally trained neural classifiers. We use a rotational equiangular tight frame-classifier and pseudo-labeling to pre-train deep encoders with few labeled samples. Furthermore, to effectively capture temporal dynamics while enforcing embedding separability, we integrate generative pretext tasks with our method, and we define a novel sequential augmentation strategy. We show that our method significantly outperforms previous pretext tasks when applied to LSTMs, transformers, and state-space models on three multivariate time series classification datasets. These results highlight the benefit of aligning pre-training objectives with theoretically grounded embedding geometry.
format Preprint
id arxiv_https___arxiv_org_abs_2508_10147
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle rETF-semiSL: Semi-Supervised Learning for Neural Collapse in Temporal Data
Xie, Yuhan
Cappelletti, William
Shoaran, Mahsa
Frossard, Pascal
Machine Learning
Artificial Intelligence
68T07
Deep neural networks for time series must capture complex temporal patterns, to effectively represent dynamic data. Self- and semi-supervised learning methods show promising results in pre-training large models, which -- when finetuned for classification -- often outperform their counterparts trained from scratch. Still, the choice of pretext training tasks is often heuristic and their transferability to downstream classification is not granted, thus we propose a novel semi-supervised pre-training strategy to enforce latent representations that satisfy the Neural Collapse phenomenon observed in optimally trained neural classifiers. We use a rotational equiangular tight frame-classifier and pseudo-labeling to pre-train deep encoders with few labeled samples. Furthermore, to effectively capture temporal dynamics while enforcing embedding separability, we integrate generative pretext tasks with our method, and we define a novel sequential augmentation strategy. We show that our method significantly outperforms previous pretext tasks when applied to LSTMs, transformers, and state-space models on three multivariate time series classification datasets. These results highlight the benefit of aligning pre-training objectives with theoretically grounded embedding geometry.
title rETF-semiSL: Semi-Supervised Learning for Neural Collapse in Temporal Data
topic Machine Learning
Artificial Intelligence
68T07
url https://arxiv.org/abs/2508.10147