Is clustering subsequences of time-series still meaningless with unsupervised learning?

Question

In the paper "Clustering of Time Series Subsequences Is Meaningless" Keoh et al. claim that breaking a time-series into chunks (sometimes called lags) of fixed-size using the rolling window method provides meaningless clustering results. From what I understand of the paper, this is because the lags are more similar to each-other than to the clusters found.

Does this problem still apply if an unsupervised method of dimension-reduction, such as tSNE, before the clustering is performed?

Answer 1

tSNE tries to place points closer to each other if they were more similar in the input domain. So it would place lags from the same series close to each other... It would likely make the problem even worse by emphasizing the nearest neighbors.