[GH-ISSUE #303] Readers should be warned that tensor decompositions are not unique #1350

New Issue

GiteaMirror · 2026-04-11T07:44:22-05:00

GiteaMirror commented

2026-04-11 07:44:22 -05:00

Originally created by @FinAminToastCrunch on GitHub (Jul 2, 2024).
Original GitHub issue: https://github.com/harvard-edge/cs249r_book/issues/303

This is regarding the portion on Tensor Train Decompositions:

https://harvard-edge.github.io/cs249r_book/contents/optimizations/optimizations.html#tensor-decomposition

I am recommending a change to this portion of the book. Tensor decompositions are not unique [1]; this can lead to confusion in the situation readers work in teams which use different hardware. I have encountered this precise conundrum in my research groups. I suggest putting a small note/footnote which lets readers know that performing a decomposition will not necessarily produce the same output across all hardware.

[1] Tensor Decomposition: Basics, Algorithms, and Recent Advances

Originally created by @FinAminToastCrunch on GitHub (Jul 2, 2024). Original GitHub issue: https://github.com/harvard-edge/cs249r_book/issues/303 This is regarding the portion on Tensor Train Decompositions: https://harvard-edge.github.io/cs249r_book/contents/optimizations/optimizations.html#tensor-decomposition I am recommending a change to this portion of the book. Tensor decompositions are not unique [1]; this can lead to confusion in the situation readers work in teams which use different hardware. I have encountered this precise conundrum in my research groups. I suggest putting a small note/footnote which lets readers know that performing a decomposition will not necessarily produce the same output across all hardware. [[1] Tensor Decomposition: Basics, Algorithms, and Recent Advances](https://link.springer.com/chapter/10.1007/978-3-031-22438-6_1)

GiteaMirror added the area: book label 2026-04-11 07:44:22 -05:00

GiteaMirror closed this issue

2026-04-11 07:44:24 -05:00

GiteaMirror commented

2026-04-11 07:44:26 -05:00

@profvjreddi commented on GitHub (Apr 3, 2026):

Hey @FinAminToastCrunch thanks for the note. With the restructure of the book into two volumes focused on ML systems infrastructure, the tensor content has been significantly reworked. Going to close this one out. If you notice anything in the current version that looks off please feel free to open a fresh issue!

@profvjreddi commented on GitHub (Apr 3, 2026): Hey @FinAminToastCrunch thanks for the note. With the restructure of the book into two volumes focused on ML systems infrastructure, the tensor content has been significantly reworked. Going to close this one out. If you notice anything in the current version that looks off please feel free to open a fresh issue!