pjreddie

12 ms·

1.
by pjreddie 7y ago
I don’t really understand how this is any different from Overfeat or SSD...
2.
by pjreddie 9y ago
Yeah, they don't show up with a zero, it just says not enough data to show so hard to draw any conclusions. I agree this is an imperfect proxy and it's not definitive but I'd still absolutely take you up on that bet (conditio
3.
by pjreddie 9y ago
I'll take you up on that bet. http://imgur.com/a/SqCKL
4.
by pjreddie 9y ago
> accusations against huge swaths of people that are totally unsupported. The search data is meant to support the accusations. A main point is that other methods of supporting such statements (like surveys) are unreliable. > So nobody
5.
by pjreddie 9y ago
Going back to the magpie though, they don't have a cerebral cortex at all, yet exhibit remarkably intelligent behavior (tool usage, self-recognition). So, what's up with all the smart corvids?
6.
by pjreddie 9y ago
They found two really old skeletons in Europe that are older than other skeletons. They are skeletons of a "modern"-ish human. They know because the teeth of the skeletons have a specific thing that the teeth of humans have but no
7.
by pjreddie 9y ago
I don't think so, Haiqing Zhou is not on that paper. Zhifeng Ren has a number of papers in this space.
8.
by pjreddie 9y ago
I think this is the paper: http://pubs.rsc.org/en/content/articlepdf/2016/ta/c6ta02876d Highly active and durable self-standing WS2/graphene hybrid catalysts for the hydrogen evolution reaction
9.
by pjreddie 9y ago
NIPS is more math/theory heavy as a conference while ICCV/CVPR/other vision conferences will tolerate less math and more pictures. You have probably been reading mostly vision papers but if you read more papers targeted at NI
10.
by pjreddie 9y ago
Mini-batches are not an ugly hack, batch gradient descent is too slow since you have to go through the whole data and stochastic gradient descent is too high variance (plus you can't do cool things like batch-norm). Mini-batches give y
11.
by pjreddie 10y ago
If Intel is smart they will invest heavily into library development. Researchers use whatever is fast and right now cuda is fast, not just because nvidia has the best GPUs (it's about even) but because the primitives in cuda are so muc
12.
by pjreddie 10y ago
Yes. You would have to have a large training set with these labels but it would be pretty straightforward to train. You would probably want a tagging model not a classifier because there could be multiple objects of interest in the same ima
13.
Hardware Guide: Neural Networks on GPUs(pjreddie.com)
8 points by pjreddie 11y ago|0 comments
14.
Real-Time Object Detection With YOLO(pjreddie.com)
13 points by pjreddie 11y ago|0 comments
15.
by pjreddie 12y ago
http://en.wikipedia.org/wiki/Ignaz_Semmelweis > Despite various publications of results where hand-washing reduced mortality to below 1%, Semmelweis's observations conflicted with the established scientific and
16.
by pjreddie 12y ago
Doctors rejected hand washing for a long time because there was no theory behind why it would improve patient care, even when there was overwhelming evidence showing it decreased mortality rates. Do you discount all evidence that doesn&#x27
17.
by pjreddie 12y ago
seL4 is 9000 lines of C and took 11 person-years to verify. How big is OpenSSL? Also, verification isn't a magic bullet, you need a good spec.
18.
by pjreddie 12y ago
I love that people do this. Do you really thing someone is going to write a web scraper and NOT include obvious things like converting AT -> @ and DOT com -> .com? Also, who even worries about spam any more? I haven't seen actual
19.
by pjreddie 12y ago
Bittorrent Sync is an amazing product, I'd recommend switching even if you don't have political issues with dropbox. I use it to keep work files in sync between 3 different computers as well as to backup 100s of GBs of pictures, s
20.
by pjreddie 13y ago
That's interesting, I don't generally think to myself, "Hm, I could go drive around for an hour or I could go for an hour long walk", I usually think "Hm, I could drive to this place or bike over there (I don't usually walk)". Although I al
21.
by pjreddie 13y ago
Well walking is way more dangerous per mile than driving, this was touched on by Steve Levitt in Super Freakonomics, it looks like in Britain the mortality rate is about 10 times higher per vehicle mile http://road.cc/content/news/68212-df
22.
by pjreddie 13y ago
Real world currencies are valued solely on the desire to have them; GDP and credit worthiness only play a role because they affect people's perception of how stable or valuable a currency is. There is nothing tangible tying the dollar to th
23.
Big Data, Big Brother and the 'Like' Button(psmag.com)
2 points by pjreddie 14y ago|0 comments
24.
by pjreddie 14y ago
At my last job I worked at a YC startup where the entire codebase was written in the 3 months leading up to demo day when the team was under a lot of time pressure. Sure it worked fine as a demo, but with the pace of startup land this same