Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
I'm grading some district reporting requirement art tests. Each art teacher has an identical test, and grades about 300 of them a year. The current iteration is marker, crayon, pencil, and paper, where we grade them by hand, then manually enter the number for each question into a database. It is horribly tedious, because grading them requires judgement and ambiguity (is that doodle textural? It's not very good texture, but they aren't very good at drawing...), but we aren't actually learning anything very useful from them, and they aren't actually aligned with the national standards. The other teachers and district office are open to a different approach, as long as it produces a numerical score and is less useless than the current iteration. The kids have Chromebooks they could bring, if needed.
Ideally, we would be assessing the national art standards (students can come up with an idea, produce something using that idea, connect it to some existing art, and articulate what they made the choices they did), but we haven't been able to figure out a way to assess that in half an hour or so and grade it in about two minutes, so currently we're assessing elements of art and a few common concepts (line, color, texture, symmetry, variety, geometric vs organic shapes).
It there a way to design an automated test in the future, which isn't primarily a reading comprehension test?
Do you have (lossless) digital copies of graded artworks from the past? If so you can train a CNN to output the grades, something along those lines. Deep Learning seems to be the obvious answer to automating hard-to-systemize yet repetitive work. It doesn't even have to be low error, you can do a last-second QC on the final grades.
If you want something more involved, you could conjure up a system where conceptual similarity could be assessed using some sort of distance metric in the latent space of stable diffusion, quality could be determined as before, and other things could also be extracted from the latent space if enough thought is put into it.
I am aware that this is far outside the budget (might have to hire a ML engineer or two to build and maintain the software) and Overtons window of public education. So individually you might just try to apply some sort of assembly line technique for your own set of artworks to grade. Give all the artworks a line grade, then a color grade, then a texture grade,... etc, and finally just sum up the weighted averages using excel or something; You might be fucking over some students whose artworks are greater than the sum of their parts or whatever (you can add a " ceiling function holistic multiplier" to counteract this ), but definitely faster than grading every piece "holistically" one by one, should be fairly statistically reliable as well. This assembly line system is the goto for where you need to sort through piles of candidates such as graduate schools or job openings (and fucks over "greater than the sum of the parts" candidates all the time).
Thanks!
Yeah, that sounds way too advanced for us, we could probably make a lot more money if we had those kinds of skills.
The second proposal might be possible -- my mother said that at her school they were given paid PLC time to do that with a test, and we spent three hours arguing about the test already this year. I we had assembly line graded them instead, they would be graded already by now.
These tests don't have to effect the kids' grades, so I don't think it will be problem from the kids' perspective.
Ideally, we would have a school approved app where we could upload a picture of a landscape, and say "Find the horizon." Maybe more than one for multiple versions. We would mark an area close enough to the horizon to qualify as right. Then it would show another picture: organic shape. We would designate which parts were the organic shapes, they would click, and it would grade based on if they clicked in one of the specified areas. Something like that. This seems like it should exist, but this is the only test I'm involved in making, so don't know for sure.
This sounds concrete enough I might be able to ask the educational technologist about it, which might be a lead, anyway.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link