karpathy

22 ms·

181.
Neuroscience for the People: Bring DIY Science to the World(kickstarter.com)
52 points by karpathy 13y ago|4 comments
182.
by karpathy 13y ago
These are usually implemented using Structure from Motion techniques, but more specifically in this case SLAM (Simultaneous Localization and Mapping). There are two sources of information: from vision and from the Inertial Measurement Units
183.
by karpathy 13y ago
This was my first reaction as well. Blocking methylation is a drastically global operation for a cell as methylation is involved in so much of the epigenetic regulation. Turning it off probably disabled much of the normal functioning of tho
184.
by karpathy 13y ago
82 participants. 3 participants did not complete. 1 person excluded as inconsistent > 4SD. Study comes down to: Participants were text-messaged 5 times per day between 10am and midnight over 14-days. """ Each text-message
185.
by karpathy 13y ago
I was hoping that the article would also mention some approximate amounts you have to drink per day to start to get withdrawal effects. Are there no studies of this? As one personal data point, I drink 2-4 cups daily and a few months ago I
186.
by karpathy 13y ago
I thought the same thing exactly. I usually convert everything on molecular scales in terms of proportions to the nearest tangible entity. (I find nm too abstract and harder to remember) A single hydrogen atom has diameter of about 10e-10 m
187.
by karpathy 13y ago
One part that's currently missing from Google Glass and that sets it apart from the dashcam example is the ability to continuously record on a circular buffer, mostly due to constraints in battery life. But once that's possible I
188.
by karpathy 13y ago
This is great write up, thanks for the explanation. I've studied physics in undergrad but I still made the mistake you describe in your last paragraph. The Grasshopper intuitively feels dynamically like an inverted pendulum but even ha
189.
by karpathy 13y ago
(Also equivalently and more commonly known as the Ising Model). The D-wave chip is just a very fast hardware-based Ising model solver. You can use it to both find the minimum of a problem through cooling, or by controlling the temperature o
190.
by karpathy 13y ago
This is awesome. I will take [a single huge list of self-explanatory example snippets (with a few short comments here and there) for 95% of use cases] over [detailed walls-of-text docs pages spread across a hierarchy of topics] any day.
191.
by karpathy 13y ago
I only found out After they made me watch their 30 second commercial. Classy.
192.
by karpathy 14y ago
I was also fairly disappointed, it looks like 5 out of 5 of my ideas for what I wanted to do with Glass can't be done with this API. But I'm hoping they wanted to get some basic / easier to implement API out as fast as possible while still
193.
by karpathy 14y ago
Hey everyone, I put this site together. It looks like the traffic managed to crash the entire cs.stanford.edu . (I hope I'm not in too much trouble...) At least for now though, I've put up a mirror here: http://badmephisto.com/glass/ plea
194.
by karpathy 14y ago
For anyone interested, 23andme is also starting to sequence exomes (the important parts of the protein coding regions) for $999, which can potentially provide much more interesting information than simply looking at SNPs. But from the websi
195.
by karpathy 14y ago
I used to be a huge fan of Prismatic until it mysteriously decided about a month ago that all I want to hear about on main feed is Apple, Gadgets and Social Media. My other 63 interests (including Physics/Web dev/Genomics/Machine Learning/e
196.
by karpathy 14y ago
Agreed with this, most notable example of this for me was when they took away recommendations within Google Reader. That was a great place to go to see what blog posts my friends like. Now it's all multiplexed into their G+ timelines and I
197.
by karpathy 14y ago
"over the years usage has declined" That's because there was almost ZERO innovation done on this product. Very few (mostly visual) improvements, and very few new features. The latest posts on Google Reader blog are from 2011 http://googler
198.
by karpathy 14y ago
The problem with this type of research in particular is that these deep learning networks shine when they are trained for a very long time across many machines, and on huge quantities of data. In other words, serious engineering muscle is n
199.
by karpathy 14y ago
Just in case someone is wondering, we are not even remotely close to technology like this. The control problem is very hard, but the perception problem perhaps even harder, Especially when most of the processing needs to be done onboard be
200.
by karpathy 14y ago
It's interesting that picture sharing is so popular and ubiquitous and yet the problem of sharing the natural extension, videos or clips, has proven to be surprisingly elusive over the last few years. What is the missing component? I think
201.
by karpathy 14y ago
I was thinking the same things, but I just took about half hour to map out my books/tvshows/movies/sports/activities and enter them all because they are now potentially searchable. For example, if one of my friends also read a book I liked,
202.
by karpathy 14y ago
Imo, this is a very smart move if the execution lives up. It gives an incentive to put up even more (structured) information to Facebook, such as Likes/people tags/etc., because now there is a more permanent point to it: it can be discovere
203.
California to Give Web Courses a Big Trial(nytimes.com)
5 points by karpathy 14y ago|0 comments
204.
by karpathy 14y ago
I can really relate to a lot of this. I'm a graduate student with ~10 distinct projects/ideas on my todo list and at any point in time I feel more or less inclined to work on each of them. And strangely, the most important ones are rarely o
205.
by karpathy 14y ago
My preferred way to think about it is in linear algebra terms: your signal of length n is an n-dimensional vector. Now you just need to change the coordinate system into fourier basis, which is made up of vectors of length n who's entries a
206.
by karpathy 14y ago
doh, not very carefully worded now that I'm re-reading my answer, you're right of course. Well, at least we're slowly converging on the right answer over several comments :)
207.
by karpathy 14y ago
Closer, but still no :) Geoff Hinton proposed contrastive divergence training for Restricted Boltzmann Machines in his 2006 science paper. CD does not apply outside of RBMs though, and most of these nets in the article here are not in fact
208.
by karpathy 14y ago
I know, right? How is this missing? For Sublime on Ubuntu: Create folder NewThemes in ~/.config/sublime-text-2/Packages/NewThemes and dump the files from the repository in there. Then Preferences > Color Scheme > NewThemes > [ch
209.
by karpathy 14y ago
Maybe this is a silly question but why doesn't SpaceX or Tesla use some form of crowd-funding scheme for their projects in addition to all their other sources of funding? I really want both companies to succeed and I'd be happy to contribut
210.
ACE to Evaluate Credit Equivalency for Coursera’s Online Courses(blog.coursera.org)
3 points by karpathy 14y ago|0 comments

More ›