Temporally coherent visualisation of time-dependent data

Alexander Modell

Keywords: visualisation, t-SNE, dimension reduction, dynamic data, temporal data

TL;DR: We propose a new variant of the t-SNE algorithm for time-dependent data which promotes temporal coherence.

Abstract: Dimension reduction algorithms aim to embed high-dimensional datasets into a low-dimensional space in such a way that important structural properties, such as clusters and manifolds, are preserved. Most such methods are designed for static data, and naively applying them to time-dependent data can lead to unstable embeddings which do not meaningfully capture the temporal evolution of the data. In this paper, we propose a new variant of the t-SNE algorithm for time-dependent data, TC-tSNE (Temporally Coherent t-SNE) in which an extra term is added to the cost function to promote temporal coherence: the notion that a data point which has a similar position in two time frames should be embedded to similar positions at those times. Importantly, this notion captures temporal similarities over the entire time domain and can therefore capture long-range temporal patterns, not just local ones. We demonstrate the effectiveness of our method for visualising dynamic network embedding, and we evaluate our method on six benchmark datasets using a collection of metrics, which capture the structural quality and the temporal coherence of the embeddings. We compare our method with existing dynamic visualisation algorithms and find that it performs competitively.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 10160