blt

81 ms·

931.
by blt 9y ago
Latex with careful tuning of line spacing, column width, etc. I don't like resumes that are too slick looking. It makes me feel like the applicant is trying to hide something. On other hand, a truly ugly resume hurts the applicant. The
932.
by blt 9y ago
I tried to find the source of the quote on epilepsy from Hippocrates. It appears on Goodreads [1] but not in the version on classics.mit.edu [2]. A different version can be found at [3], but it also does not match. It seems that people are
933.
by blt 9y ago
Yes, but this paper is all about incorporating prior knowledge of human locomotion in a corrective step. There is no innovation in the field of pure inertial navigation; this would not help e.g. a quadrotor.
934.
by blt 9y ago
Thanks for sharing this!
935.
by blt 9y ago
What is novel about that? Phones have had them for years
936.
by blt 9y ago
Dead reckoning is like pure IMU integration. In this work they detect the instant of zero velocity and do something like a Kalman update with a velocity measurement
937.
A Global Optimization Algorithm Worth Using(blog.dlib.net)
265 points by blt 9y ago|63 comments
938.
by blt 9y ago
You mean tasks like computing the rank of a matrix? ( http://oneweirdkerneltrick.com/rank.pdf ) ;) Image super-resolution is an ill-posed inverse problem: many possible high-resolution images would be reduced to the same low-
939.
by blt 9y ago
good point!
940.
by blt 9y ago
waifu2x ( http://waifu2x.udp.jp/ ) is another often-posted superresolution approach, focusing on upscaling low-res anime images instead of extremely tiny pixel art. It uses a totally different learning-based approach - howeve
941.
by blt 9y ago
Tell me about it, I'm using tensorflow now and it does amazing things, but the API is so bulky for anything besides setting up the computational graph... Planning to check out pytorch for next project. Imo most dev environments and sof
942.
by blt 9y ago
There was a good talk at NIPS this year about the difficulty of reproducing results and benchmarking in RL. In supervised learning you can easily compare results on standard datasets. In RL though, once the actions of two policies diverge,
943.
by blt 9y ago
Saving movies to disk doesn't make much sense. The files are huge, and rewatching is not that common. BitTorrent is great for music, sadly we lost what.cd to law enforcement.
944.
by blt 9y ago
I hope this faculty shortage is still in effect when I finish my Ph.D. :) Anecdotally, many of my peers are going to industry instead of academia, because 1) better pay, 2) no begging for funding, 3) they don't want to teach. Academia
945.
by blt 9y ago
> Pointing out someone's implicit biases is not necessarily a personal attack on them. Everyone has them so we should welcome when people point them out (unless done in an mean-spirited or aggressive fashion). I wish I could upvote
946.
by blt 9y ago
I agree about the software crisis, I guess I think the solution will come from a tool set that is not yet (widely) known.
947.
by blt 9y ago
IMO, there was big churn 1990-2015 in people's attitudes about the right way to develop software. This manifested in lots of discussion about languages, platforms, program architectures, hardware, etc. Since I started reading HN around
948.
by blt 9y ago
is there source code to SC2K anywhere? If not, won't it be really hard to even approximate the simulation behavior?
949.
by blt 9y ago
Yeah I would detect the cuts with something else. I just need the library to read/write video files, display them on screen, etc.
950.
by blt 9y ago
Sounds great, but I still have trouble with the idea of a $10,000 machine that is permanently married to one monitor and unexpandable. If I'm spending that much on a computer I want PCI slots. Especially now that it's common to ke
951.
by blt 9y ago
This looks cool. I am planning to write some tools for semi-automated processing of a video dataset soon. Are there any other strong alternatives to this library that I should consider? Am OK with any language. An example task: automaticall
952.
by blt 9y ago
I don't understand how it took an entire book to show that a linear classifier can't learn the xor function.
953.
by blt 9y ago
What's your point? Yes, those two goals are in conflict. Modules are still an improvement to the current situation with no downsides.
954.
by blt 9y ago
Is "tangency" really a widespread term for curvature discontinuity?
955.
by blt 9y ago
It's convoluted because the idea of forcing an object to always be allocated on the heap is convoluted. The real fix for this problem is the modules system, where the compiled module can expose an object's size without exposing it
956.
by blt 9y ago
I would always pick a make file over a script when it's appropriate - why did you shy away from them?
957.
by blt 9y ago
Sounds like a sweet spot kind of project for C++.
958.
by blt 9y ago
The truck driver thing is not true: https://www.marketwatch.com/story/no-truck-driver-isnt-the-m...
959.
by blt 9y ago
While matrix derivatives are important, there is also a lot of other math in DL papers. In particular, a lot of the probability side concerns expectations, KL divergences, entropy, etc., which are all defined in terms of integrals or sums.
960.
by blt 9y ago
The book "Probabilistic Robotics" explains the Kalman filter really well by putting it in the larger context as a Bayesian filter. I understood it only after reading that book

More ›