imurray

16 ms·

121.
by imurray 10y ago
Autograders are a mixed blessing. At my University some of our courses have autograders, and I think they can be great. There is a compiler course where the main goal is to create a compiler that has a specific set of measurable features. E
122.
by imurray 10y ago
If you mainly care about prediction, rather than inspecting the fitted parameters, a lot of this detail is usually overkill. To generalize well, it's almost always a good idea to have some sort of regularization, such as penalizing the
123.
by imurray 10y ago
In machine learning "evaluating the gradient" means sweeping over the whole training set. For simple models, stochastic gradient descent will have found a good fit after one or two passes over the dataset. (For neural nets, tens t
124.
by imurray 10y ago
I agree. But part of it is sociology: people use what their friends use.
125.
by imurray 10y ago
For a mathy academic survey of gradient descent methods (from this month, June 2016), I recommend: https://arxiv.org/abs/1606.04838 (It was submitted by someone else a few days ago, but saw no traction.)
126.
by imurray 10y ago
BFGS is an example of a quasi-Newton method. L-BFGS is the popular "low-memory" variant. Any such method can be used with the two references I gave above. One could also attempt to approximate each of its internal computations to
127.
by imurray 10y ago
Stochastic gradient descent methods are the most-used for neural net fitting, so the most discussed. However, the methods you mention are sometimes used too. For the closest thing to Newton's method applied to neural nets, search for &
128.
by imurray 11y ago
If you have libtermkey.so.1 somewhere, you could export LD_LIBRARY_PATH to point to the directory containing it. More notes on what to do if a build system for a project doesn't link in libraries, and you can't simply install the
129.
by imurray 11y ago
David MacKay's landing page: http://www.inference.phy.cam.ac.uk/mackay/itila/ Other formats are available along with some other information. David's course and drafts of this book were my introduction to
130.
by imurray 11y ago
Reminds me of the following different project for content aware image resizing or retargeting: https://www.youtube.com/watch?v=qadw0BRKeMk https://en.wikipedia.org/wiki/Seam_carving
131.
by imurray 11y ago
A kerning game: http://type.method.ac/
132.
by imurray 11y ago
This was the most attended of 6 tutorials at NIPS 2015. All of the tutorials are listed at https://nips.cc/Conferences/2015/Schedule?type=Tutorial where slides are available now, and video will be when the Microso
133.
by imurray 11y ago
Part of me wants to just hand-write them and then maybe scan them, but that's not searchable/indexable, and doesn't lend itself well to version control. I'll comment as someone who does hand-write and scan my notes. I
134.
by imurray 11y ago
I'm well aware of how and when it works. It is the recognized idiom to perform in-place operations in practice in Matlab code [1]. It would be nice to take advantage of the same idea in mex code where possible. For example an API funct
135.
by imurray 11y ago
Here's an example: A = plus_diag(A, x); % Using [1] below Adds a vector x onto the diagonal elements of a huge matrix A. In old versions of Matlab, 2*huge memory would temporarily be allocated. Now the memory usage doesn&#x27
136.
by imurray 11y ago
A different (probably slightly cheaper) hack to get a safe copy of an array is: A = X; A(1) = A(1); Then one can evilly edit A in-place in a mex file without accidentally editing other arrays like X. However, the docs tell us never
137.
by imurray 12y ago
The corresponding paper was posted recently, and there were a lot of comments: https://news.ycombinator.com/item?id=9202039
138.
by imurray 12y ago
[Assuming you meant LDA = Linear Discriminant Analysis.] It looks like the toolkit implements "unsupervised" methods, where the datapoints don't have a special "label" feature that is treated specially in the embedd
139.
by imurray 12y ago
One can try to catch use of gcc extensions with: gcc -std=c99 -pedantic
140.
by imurray 12y ago
Hopefully not (although possibly) just like Skype: https://en.wikipedia.org/wiki/Skype_security#Eavesdropping_b... Presumably the encryption of open source SIP clients and Firefox hello are more trustworthy(?).
141.
by imurray 12y ago
This warning is spot on. "Since it's odorless and colorless, you probably wouldn't even notice." — I remember being surprised reading a haz-card for liquid nitrogen that said you could suffer serious brain damage from
142.
by imurray 12y ago
Wikipedia tells me that lasers have been considered before to fight off the https://en.wikipedia.org/wiki/Kessler_syndrome
143.
by imurray 12y ago
There are lots of ſ's and other slightly hard-to-read typesetting when reading papers from the Royal Society. It's freely-available and has some quaint gems: https://royalsociety.org/news/2011/Royal-Soci
144.
by imurray 12y ago
It would be safest for most rand() functions to omit both zero and one, unless a user was really sure they wanted otherwise. If we were generating real numbers, we'd never see precisely zero or one. The fact that we do is an artifa
145.
by imurray 12y ago
Here's a dump of some other insights into the referendum: Betfair's probability of a majority yes vote (NOT a poll of fraction who will vote yes) over time: http://uk.sportsiteexweb.betfair.com/betting/LoadRun
146.
Does Earth really have two high-tide bulges on opposite sides?(physics.stackexchange.com)
2 points by imurray 12y ago|0 comments
147.
by imurray 12y ago
Reminds me of Averia, a font made by averaging fonts. http://iotic.com/averia/ https://news.ycombinator.com/item?id=3877025
148.
by imurray 12y ago
I briefly wondered how to get push notifications from a Google spreadsheet, to update a website. My guess is they do it by getting email notifications: https://support.google.com/docs/answer/91588?hl=en
149.
by imurray 12y ago
The intention (3rd link) was for papers and discussion. There were probably more comments in the first few days before interest fizzled. Gaining critical mass is hard.
150.
by imurray 12y ago
Someone did briefly run an "Academic Hacker News" about 5 years ago (on HN's arc software). https://news.ycombinator.com/item?id=506064 http://web.archive.org/web/20100107015442/http

More ›