dimatura

33 ms·

361.
by dimatura 9y ago
As long as you're citing @karpathy, "I've been using PyTorch a few months now and I've never felt better. I have more energy. My skin is clearer. My eye sight has improved." ( https://twitter.com/karp
362.
by dimatura 9y ago
That actually happened a long time ago.
363.
by dimatura 10y ago
I'm sympathetic to the idea that robotics should try to do more with cameras, as opposed to LiDAR. As you say, humans can do without LiDAR (and they don't even need stereo!). But in practice, there's a good case for LiDAR in
364.
by dimatura 10y ago
Sweet!
365.
by dimatura 10y ago
Is ARM, and in particular the NVIDIA TK1/TX1/TX2 supported? Ease of use on these platforms with Theano vs Tensorflow/PyTorch is the main thing keeping me on Theano.
366.
by dimatura 10y ago
It does not, at least not explicitly, which is a distinguishing feature of policy optimization algorithms. I don't think there's anything about this that makes it worse for non-deterministic problems. Note that the policy can stil
367.
by dimatura 10y ago
While they're in the same general family of black box optimization algorithms, ES and GAs are not really the same, and in fact were developed independently. The historical roots of both are actually quite interesting.
368.
by dimatura 10y ago
As a grad student in the field, I 100% agree with the idea of research debt. Arxiv is great, but it has only exacerbated the problem. There's far too many papers adding some small fluorish on an already complex architecture in order to
369.
by dimatura 10y ago
Agreed. It's worth noting that Arduinos, by default, are programmed in C++ (thanks the AVR port of GCC), and it's been this way more or less since their first version, which (as far as I recall) was a fairly small 8-bit micro. The
370.
by dimatura 10y ago
Doesn't libigl use Eigen for matrices? Eigen is a full-featured matrix library with a pretty decent API. If you just need a 4x4 grid of numbers, then you don't really need a library, sure. But once you need to do matrix arithmeti
371.
by dimatura 10y ago
I've used BlaBlaCar in Germany. They have buses operating in a similar niche as Greyhound/Megabus (cheap but slow). Compared to those, the BlaBlaCar option was maybe slightly more expensive but faster and more comfortable.
372.
by dimatura 10y ago
I used BlaBlaCar a couple of times to travel between Berlin and Hamburg. It was great, and I've wished there was an equivalent here in the US for traveling between NYC/Philly/Boston/etc. In both cases I traveled with peo
373.
by dimatura 10y ago
This is great! The network visualizations are also pretty sweet. How are those generated?
374.
by dimatura 11y ago
In my experience the Tegra K1 is rather limited as a deep learning platform. A Tegra X1 would easily fit in this platform, and is about twice as powerful (at about twice the cost, admittedly). In fact, for a slow-moving ground robot such as
375.
by dimatura 11y ago
Backprop may be added in the future. Meanwhile you can just adjust the weights directly! (They're in the "layer*" sheets).
376.
by dimatura 11y ago
I think it's feasible, even without macros! Just have to make our case to our advisors for it.
377.
by dimatura 11y ago
Oh boy. I'm the second author. The official announcement will be in April 1st at the SIGBOVIK conference ( http://sigbovik.org/2016 ), but I guess someone broke the press embargo! By the way: It works, try it ;)
378.
by dimatura 11y ago
I agree VSLAM is good and getting better everyday, but I don't necessarily agree it'll displace LiDAR for autonomous vehicles. The first generations will probably have both. Sure, LiDAR is more expensive, but we're talking ab
379.
by dimatura 11y ago
You can get a Velodyne "puck", that is light enough for many drones, for around $7k. The problem is that the scan density is not that great, so for many applications (e.g. archaeology) I suspect it's not worth it. Structure f
380.
by dimatura 11y ago
> you don't have to do object recognition on the point cloud This is true, as long as you're willing to treat everything your LiDAR sees equally. The problem with this is that you're treating, say, a patch of grass (which
381.
by dimatura 11y ago
I use mainly two tools: The vim notes plugin ( https://github.com/xolox/vim-notes ) and zim wiki ( http://zim-wiki.org ). vim notes is very good at quickly jotting down and bringing up things from vim; much bet
382.
by dimatura 11y ago
I also love the thinkpad trackpoint, and I got one of the keyboards as well. But I can't deal with the control/fn key swap; I swap it in my laptop's bios, but can't do it on the keyboard. And I use ratpoison with vim bin
383.
by dimatura 11y ago
Our life is dominated by systems we don't understand. I have some understanding of how my cell phone works at the software level, but when it comes to details at the hardware level I just trust the electrical engineers knew what they&#
384.
by dimatura 12y ago
I have the same problem. I usually end up bookmarking to pinboard instead, where I can add tags. Also, I often find bookmarking at the repo level too coarse - often I am only interested in a specific file or class.
385.
by dimatura 12y ago
This is something addressed by probability theory, not CS. While MCMC is easier to explain and understand for discrete spaces, it can also be used for continuous spaces. As a non-mathematician I just think about this as "discretizing a
386.
by dimatura 12y ago
In theory, at least, a deep architecture can learn the same things as a shallow one, but with less nodes. See http://www.iro.umontreal.ca/~lisa/publications2/index.php/pu...
387.
by dimatura 12y ago
You can always clone/fork the repositories of the plugins you're using. Having moved from a big repo for my .vim + .vimrc to Vundle, I appreciate the possibility of quickly adding and removing packages.
388.
by dimatura 12y ago
I would also suggest K. Murphy's Machine Learning for the journeyman level. In the intermediate apprentice-journeyman level Alpaydin's Introduction to Machine Learning is very friendly.
389.
by dimatura 12y ago
This is very much the same reasoning behind protothreads[1]. As a bonus, the implementation of protothreads uses a neat trick reminiscent of Duff's device [2]. [1] http://dunkels.com/adam/pt/ [2] http:/
390.
by dimatura 13y ago
When I was at CSAIL, I got one of those targeted phishing emails like "YOURE MIT INBOX IS FULL!!11!!" full of typos and painfully fake. I laughed it off, thinking MIT CS students would be too smart for this, but I was surprised it

More ›