mbq

20 ms·

211.
by mbq 13y ago
The difference between statistical modelling and machine learning is that the first aims to make models that have a "right" accuracy on a given data-model complexity balance while the second those that work well on "unseen da
212.
RSS feeds with all DBLP-indexed Computer Science papers(services.ceon.pl)
1 points by mbq 13y ago|0 comments
213.
by mbq 13y ago
Nope, that Eve can intercept a copy of GPG-encrypted mail and extract plaintext To, From, Title, time, with some luck also some info about sender's IP and software, plus all info that is in recipient's and possibly sender's p
214.
by mbq 13y ago
Something a bit different, but worth mentioning: http://barada.sourceforge.net/
215.
by mbq 13y ago
There is something called ARFF which is basically a CSV with a header that defines how to interpret data, a bit of documentation and a reference implementation.
216.
by mbq 13y ago
Oh sure, let's keep buffer bugs, waste tons of bandwidth and storage, burn millions of CPU hours in parsing computer-generated text, completely ban encryption and screw it for all non-en_US-speakers so that it could be debugged via mis
217.
by mbq 13y ago
To be honest I don't expect the results to change; but this is an only way to attach significance to the observed differences and to ensure this wasn't a lucky shot.
218.
by mbq 13y ago
You are just doing a simple validation on a test set rather than cross-validation; the point of CV is to make many iterations of validation on different train-test splits and average the results.
219.
by mbq 13y ago
Systemd helps a little because its service files are declarative and thus easier to understand and write; also stuff like detecting processes or managing commands are done in a consistent way.
220.
by mbq 13y ago
You can start your own mail server and make it encrypt all incoming messages with your GPG or SMIME public key. This way NSA will have to collect your old messages from various senders' servers (or the archive made by their sniffers).
221.
by mbq 13y ago
This is precisely why I think modern approach to software UX and design is crazy. Like 15 years ago, in the Windows hegemony every sane designer was heavily relying on widgets, metaphors and flow defined by the system and MS could almost fl
222.
by mbq 13y ago
Two more questions then; is this speed-up achieved on a single core or also due to parallelism? And have you compared WiseRF with Random Jungle ( http://bioinformatics.oxfordjournals.org/content/26/14/1752.... )?
223.
by mbq 13y ago
It's not that bad, moreover the game is always about quality/quantity ratio. And if it had become a standard, its development would be heavily pushed forward and vp8 would likely now be really free and called theora 2.0.
224.
by mbq 13y ago
Ogg as a fallback for Linux... Actually Theora was seriously planned to be the codec for HTML5 video, but Apple and Nokia spoiled this in 2007 due to some virtual patent problems.
225.
by mbq 13y ago
There are many problems with the golden model (like really golden prices if you compare them to their real costs), but it is not true that it implies bad quality -- greed does. Springer's BMC for instance is golden and quite reliable, while
226.
by mbq 13y ago
Comparison of frames? I was only thinking about averaging predictions for each frame which is just 20x slower. Still I think that the process is way more constrained by the time to fetch new CAPTCHAs than by OCR, so this doesn't really matt
227.
by mbq 13y ago
Also the funny part is that there are many nice ML algos that do not need harsh optimisations or even do not need optimisation at all (see random ferns for instance). Thus it might be hype for hype.
228.
Why are there exactly four nucleobases in DNA?(biology.stackexchange.com)
2 points by mbq 13y ago|0 comments
229.
by mbq 13y ago
It can look at all frames and average predictions; this should boost the accuracy as errors will likely cancel out.
230.
by mbq 13y ago
A minute of googling and you can find a direct download from other databases, for instance: http://kmoddl.library.cornell.edu/bib.php?m=15 . But yes, that's awful that Google is blocking this stuff -- in my country it is even entirely blo
231.
by mbq 13y ago
+1 For Luxi Mono; technically it is terrible for programming (0&O and 1&l pairs look almost the same), but I'm somehow addicted to it. So I really hope one day someone will make open source "Luxi Mono-g" clone...
232.
by mbq 13y ago
The problem is that still most data is not big; even worse, big data is often either small data presented in a very sparse way (logs for instance) or is being subsetted or summarised to fit in Excel :(
233.
by mbq 13y ago
Or use a machine learning method that is fully invariant to all monotonic transformations, for instance random forest.
234.
by mbq 14y ago
Way worse. Evolution is not about gaining complexity but about fitting to (current state of) the environment; in this manner, Procaryota are actually the most "evolved" form of life since they can rapidly adapt thanks to their short life cy
235.
by mbq 14y ago
Well, on Android you can install TextSecure and have even a more secure messaging. And replace the stock system with Cyanogen mod or similar and have a smartphone that won't be taking commands from the mothership.
236.
by mbq 14y ago
This super-forecast is rather naive; for a given location and day-of-year it is the same for any future year.
237.
by mbq 14y ago
Doesn't seem science.
238.
by mbq 14y ago
As far as socket reservation and auto-rebooting is concerned, systemd itself is a nice solution.
239.
by mbq 14y ago
Browser sandboxes are regularly escaped, both between same-origins and surroundings of the browser. Native apps are usually verified cryptographically, mobile almost never. Webapps shamelessly reveal many information about your usage patter
240.
by mbq 14y ago
Native will never die, and this is because there is a fundamental problem with webapps -- they lay on a wrong abstraction, a document delivery system. To this end, they will always require huge amount of hacks to overcome its restrictions (

More ›