tysam_and

11 ms·

61.
by tysam_and 3y ago
I have autism as well. I don't see normal people as disabled, just different. I think seeing normal people as disabled is a rather myopic perspective, and I sometimes hear it from the community. I don't like it, personally. Typica
62.
by tysam_and 3y ago
I ended up cleaning it up to share it and got sucked back into the project. >:'(((( Spent 2 hours today in RawTherapee trying to improve scan contrasts :'(((( I hope to post a perhaps more cleaned up version of at least the dat
63.
by tysam_and 3y ago
I actually participated in the challenge for a little while and this was the approach I took before I dropped out to do a few other things. However, what I did was a bit different -- instead of looking for a crackle, I surmised that that &#
64.
by tysam_and 3y ago
Thank you, very much appreciated, this means a ton. Thank you for helping me make the connection, I will look into this. <3 :'))))
65.
by tysam_and 3y ago
Yes, empirically I've noticed that if you do not exchange that sequence length gain for huge batch sizes up front, however, your performance will go down overall....
66.
by tysam_and 3y ago
If they can make a 288 GB $4.4-6.8k prosumer, home-computer-friendly graphics card, I will be extremely happy. Might be a pipe dream (today at least, lol, and standard in like...what, 5 years?), but if they can pull that off, then I think t
67.
by tysam_and 3y ago
It seems like Hotz and co are able to move pretty well on it, so maybe there's some low-level stuff they're using (or maybe they're forced to for a few reasons) w.r.t. the tinybox, but it is impressive how much they've b
68.
by tysam_and 3y ago
Gotcha, thanks. I guess it could be useful as shorthand for specifying 'joint latent space of multiple modalities', since 'two tower' is shorter and more established (though I guess it had to get established at some poin
69.
by tysam_and 3y ago
I guess this is a bit outside of microcontrollers, but it is nice to see currents going towards things like FPGAs too, for example. GPUs are great but for my usecase they also can be a bit limiting, because I'd like to do high throughp
70.
by tysam_and 3y ago
I do not want to be too much of a downer, but is there something keeping us from just using the traditional verbiage we've used in the community for years and call it "projecting into a shared latent space" (or something blan
71.
by tysam_and 3y ago
I enjoyed listening to this as an audiobook. Not a ton I remember from it, but it seemed helpful to me, and the biggest takeaway is that a conversation is building something that you two are working on together (and the implied lesson beyon
72.
by tysam_and 3y ago
I'd honestly like to write a short, inexpensive, useful coffee-table book someday. I've googled 'autism dictionary' and haven't found anything useful yet, really not sure why to be honest. It would be very nice havi
73.
by tysam_and 3y ago
I'm autistic, and it's taken me a long time to learn how to start. It's been worth it. Today, a woman and I were both looking at the same display in Hobby Lobby, both clearly looking for something. When she moved over, I used
74.
by tysam_and 3y ago
I think you may have replied to the wrong comment (I might be misunderstanding, however).
75.
by tysam_and 3y ago
I'm curious if anyone knows people who would help me bring some efficient ML project work I've done to spheres like Africa. I've visited a couple times and loved being there so much, but definitely feel extremely disjoint fro
76.
by tysam_and 3y ago
Holy cow, that's freaking crazy. Great catch, and good eye! I guess they were more valuable for me since I had 'paid' for them, and felt I needed to play them in order to maximize my value, haha!
77.
by tysam_and 3y ago
What is mind-boggling to me is that it includes an actual, physical, hardware Gameboy Advance CPU in it for when it's playing Gameboy Advanced games. I was one of the people who used the 'Ambassador Program' due to paying ~$2
78.
by tysam_and 3y ago
That's what makes it a good benchmark. It's a benchmark. Not a real world problem. That's why the traditional path has been MNIST -> CIFAR10 (optionally -> CIFAR100) -> ImageNet -> ????!?. Because it gets graduall
79.
by tysam_and 3y ago
Great link, thank you so much, lol. This line freaking killed me: https://github.com/benjamin-recht/mnist_1_pt_2/blob/c0fe96bc...
80.
by tysam_and 3y ago
Ohhh, you know what? I think this makes sense, since it's basically like making prototypes from the training set, only it's instead comparing against every training example, then averaging. Cool. Did not know that would be remotel
81.
by tysam_and 3y ago
Holy cow, I knew that MNIST was simple, but not that simple. Could you post a snippet of the code that you used to achieve this? It would be really, really nice to have a baseline to work from I'm sure, and I feel like this could be re
82.
by tysam_and 3y ago
Yeah it's just a habit for what I do, feels most comfortable for me when commenting or messaging people. We humans all have our quirks! <3 :'))))
83.
by tysam_and 3y ago
While I dislike the silliness that you're alluding to, I think you're using multiple meanings of the phrase 'AI Safety' there all lumped into one negative association. There are risks, esp in a profit-motivated capitalis
84.
by tysam_and 3y ago
Yes, unfortunately I have yet to beat that technique in wallclock convergence speed vs just using the larger network from the start. :'(((( Whoever figures out how to clearly and effectively do it consistently faster than a 'full
85.
by tysam_and 3y ago
facepalm I just realized I was talking about this in the context of MNIST -- my apologies. It would be better in a problem with horizontal symmetry like CIFAR, esp if the zipping is serialized by unwinding the 2d pixel array into 1d. One
86.
by tysam_and 3y ago
And another round of races commences! Some folks seem to have some strong ire towards OpenAI (maybe a bit less recently), but for one, they seem to do a really, really, _really_ good job at making themselves "the benchmark to beat&quot
87.
by tysam_and 3y ago
Additionally, try flipping your images and averaging the size before comparing the distance between them. I'd expect a boost of about 78% -> 84% or so, based on how this typically works as TTA.
88.
by tysam_and 3y ago
"One simple trick to beat the statistical odds...."
89.
by tysam_and 3y ago
Makes sense, considering that's not too terribly far off from what the KL divergence does
90.
by tysam_and 3y ago
If you'd like to play around with MNIST yourself, I wrote a PyTorch training implementation that gets ~99.45%+ val accuracy in <13.6 seconds on a V100, est. < 6.5 seconds on an A100. Made to be edited/run in Colab: https:&#

More ›