There needs to be some sort of organized push for visualization tools. I know, I might be bringing the proverbial owls to the proverbial Athens with saying that here, but I really do feel that if done right this could impact the course of the world like nothing else. This could be as important as idk, invention of book press or smth. Make computer "the visualization machine".
I think that one of the fundamental problems is that to be a visualization machine, you need to have easy access of the GPU and OpenGL is provides anything but. I think that shadertoy (shadertoy.com) is the thing that comes the closest but the learning curve is kinda steep.
I know that people like Alan Kay, Bret Victor or Michael Nielsen (his post was on the fp the other day
https://news.ycombinator.com/item?id=15616637) share these sentiments but this is a task bigger than a single people.
Idk what I really mean by "organized push". I'm not sure if the problem is well defined too
In deep learning, TensorBoard (https://www.tensorflow.org/get_started/summaries_and_tensorb...) works with TensorFlow and Keras to show what the model is doing. However, it ends up being more complicated/unintuitive than a YouTube video, so it's not as useful.
The problem is that this is an ad hoc solution. What I'm talking about would be some formalization of visualization (I guess kinda like grammar of graphics without the statistical aspect) so you can visualize just about anything.
I feel like visualizations rely too much on the existence of a meaningful isomorphism. That is, once a problem is visualized effectively it becomes trivial and though applicable to future similar problems the isomorphism itself is too domain specific to be generalized. It feels like trying to find an analogy that will help you find all future analogies.
Isn't that what category theory is about, on the meta level, and in the result in case of specific isomorphisms, too?
edit: at that I still have John C. Baez, Mike Stay - "Physics, Topology, Logic and Computation: A Rosetta Stone" on my reading list https://arxiv.org/abs/0903.0340
There was a big organized push for visualization and more precisely augmented visualized thinking at HARC. It’s really too bad HARC didn’t work out, but many of us are still very interested in this problem.
I agree. Visualization is often key to understanding and identifying non-trivial issues.
Here's a tool a colleague of mine made for inline "visual debugging" for e.g. computer vision, written in c++: https://github.com/lightbits/vdb. I haven't used it myself, but when he presented it I think it made a lot of sense to have these sorts of tools for processing data in real time.
I think that one of the fundamental problems is that to be a visualization machine, you need to have easy access of the GPU and OpenGL is provides anything but. I think that shadertoy (shadertoy.com) is the thing that comes the closest but the learning curve is kinda steep.
I know that people like Alan Kay, Bret Victor or Michael Nielsen (his post was on the fp the other day https://news.ycombinator.com/item?id=15616637) share these sentiments but this is a task bigger than a single people.
Idk what I really mean by "organized push". I'm not sure if the problem is well defined too