Here it matters much less than in generic LLMs though. There's no chance of test set leakage since the network is not general purpose / not trained on the internet.
>Figure 18: The Taiji-DDN exhibits a surprising similarity to the ancient Chinese philosophy
of Taiji. Records of Taiji can be traced back to the I Ching (Book of Changes) from the late 9th
century BC, often described by the quote on the left (a) that explains the universe’s generation and transformation. This description coincidentally also summarizes the generation process and the transformations in the generative space of Taiji-DDN. Moreover, the diagram (b) from the book Tom (2013) bears a closely resemblance to the tree structure of DDN’s latent fig. 1b. Therefore, we have named the DDN with K = 2 as Taiji-DDN.
Very nitpicky comment, but I personally find such things to make for a bad impression.
To be more specific, branching structures are a fairly universal idea, so the choice of relating it ancient proverbs instead of something much mundane raises an eyebrow.
Not sure about the LLM community, but it's not uncommon in other computer-science communities to assign common names to models and implementations.
Common names will always be influenced by the authors culture, so it seems unfair to exclude a name based on any individual opinion that it is or isn't a mundane choice.
Unless you would also exclude names based on their relationship to old, but culturally relevant texts in the western tradition to, e.g., the bible.
