ground truth differs

work, processing, faster rates, still stuck, and summer

until you actually do something, you don’t realize and understand the reality and actuality of hindrances and difficulties, or simplicities and workarounds. though so much automation in programming, prompting, and debugging exists, the ground reality of unifying the various subprocesses, to have a cumulative framework, to yield fruitful, desired, and ultimately usable extracted results, is still a long shot in quite some fields.

i always feel examples are better and serve a valid purpose to bring reliability, so here are some. when working with imaging modalities this summer, i realized the numerous models that are involved and utilized by numerous parties in their respective subprocessing steps before we finally receive the data; replaces, truncates, modifies, fine-tunes, and polishes various components so much that backtracking gets really infeasible and troublesome. and the sequence of steps seems almost arbitrary.

this includes the usage of unreasoned affine transformations to serve a momentary purpose, composition of such functions in an arbitrary, non-documented fashion, superimposed coordinate axes shifts, ill-co-registration of modalities, a masking strategy working for extraction in one platform but non-transferrable in another, back and forth between anisotropic and isotropic variations, unrecorded or half-filled information, imbalanced sample classes, unidentifable alignments, and more.

with that being said, this doesn't take away from the many learnings developed through the dealings, workarounds, and unknotting of the imperfections. alongside developing a good amount of mental resilience to finally debug or figure out a solution each time something goes off route. similarly, no singular entity or group is at fault either. because they built their models to serve their chosen step, which just happens to lie midway amidst a multi-step process. but others who would want or have to utilize and put together these multiple moving parts for their desired tasks suffer.

this is my second summer working with imaging modalities. where in the last session, i even found a mathematical meme come true, of a framework being underlined as trivial, but then not a single explanation or article or review or paper, or document highlighting the proof, background, or implementability of it, apart from a brief theoretical gist. not even examples or boiled-down versions. also not to forget the case of outdated or mismatched versions of frameworks and modules in the system, cluster, or model setups, as well as various permission issues & laborious authorizations, and unmanaged databases & missing information.

ps: interestingly, this stark difference in ground truth also extends to real-world phenomena and scenarios too, if we notice. you witness the mess and cover of the true difficulties and misrepresented simplicities, and segregate what is valid and biased on your individual basis. the perceived difficulties or interpreted simplicities vary from person to person, thus the non-uniform representation and miscommunication of different ideas and constructs across different circles and systems, but this includes realizing actually how doable or not it is to ask someone, seek an opportunity, get over something, the actual living environment and real surrounding of a place labeled in a certain way, reality of wanting to manage and lead people, and more.