Profile

Cover photo
Alex Borodach
Works at Startups
Attended Belarusian State University
Lives in Minsk
182 followers|128,585 views
AboutPostsPhotosYouTubeReviews

Stream

Alex Borodach

Shared publicly  - 
2
Add a comment...

Alex Borodach

Shared publicly  - 
1
Andrew Osipenko's profile photo
 
Ну што ж неплохо для затравки. Но не только этим знаменит Таиланд. Давай выкладывай
 ·  Translate
Add a comment...

Alex Borodach

Shared publicly  - 
 
English's spelling system is terrible for sure. Here is why.
 
"Dearest creature in creation
Studying English pronunciation,
I will teach you in my verse
Sounds like corpse, corps, horse and worse."

So begins Gerard Nolst Trenité's poem "The Chaos," (the italics are his) a wonderful demonstration of the incomprehensibility of English pronunciation. (Which you basically have to hear read out loud; I have a fairly solid command of the language and I can't figure out which pronunciations he meant in each place. Fortunately, the video does this nicely)

People often mistake this for the English language being difficult, but it's actually something rather different: English has a relatively simple grammar, but its writing system is quite bizarre, with written forms giving only the roughest clues to how words are pronounced. It's almost closer to logographic writing systems like Kanji than it is to ordinary alphabetic writing. (To give you a comparison, several years ago I had to spend some weeks in Poland for work; as prep for this, I drilled on the basics of the language, including the rules of pronunciation. After a few hours of study, I found that I could read any passage in written Polish out loud without difficulty, and well enough that native speakers were completely flummoxed that I didn't speak the language at all. Go ahead and try that in English; I dare you.)

There's an interesting history behind this. Part of it comes from the heavy borrowing of words from other languages into English -- "croquet" and "lingerie," for example, use French pronunciation (although in the latter case, not French meaning!) and you simply have to know that in order to pronounce them; an experienced English reader will recognize the "-quet" as being a distinctly French pattern, and use those rules, much as they will pronounce "xeriscape" using a Greek-derived sound pattern, even if they don't know Greek.

But there's more to it than this, because every language that's spoken by people who travel or trade is rife with borrowing. What's unusual with English is that it hasn't had any deep spelling reforms (which is likely tied to the Anglophone world's lack of strong monarchies or other forces which imposed such reforms elsewhere) in a while: instead, there have been several waves of standardization which basically fossilized whatever was in the language at the time.

For example, one of the huge standardizing influences was the rapid rise of printing starting in the late 15th century, which caused texts to be much more widely available than ever before. With so much reading going on, people tended towards single spellings for single words. However, the period from the 14th through 18th centuries was also the time of the "Great Vowel Shift," a major transition in the way English was pronounced. Spellings which made perfect sense prior to the shift -- for example, "name" being pronounced "NAH-meh" -- suddenly made no sense at all as the vowels moved.

But at the same time that this was happening, words were getting imported into the language in tremendous bulk, and those words came from languages which were now standardizing their spelling in their own way, so the words got both the spelling and pronunciation of their parent language; thus we get, for example, "corps," which came into English from French in the late 16th century, versus "corpse," which came in via Old French several hundred years earlier.

"Corpse" was standardized with printing, being spelled (just to make this more confusing) "corps," and pronounced much like the modern word "corpse." (This word didn't happen to get modified much in the Great Vowel Shift) The French word, meanwhile, had changed a good deal under French's own sound shifts: since the Gallic "r" is very far back in the mouth, the following "ps" sound got lost. Under French's own standardization-by-printing, it kept its spelling but changed its sound, and so when English borrowed the word yet again, it kept the spelling "corps" and the French pronunciation "kor," and a final "e" was added to "corpse" to distinguish them -- probably on the theory that silent "e" was a fairly common feature at this point (thanks to the Great Vowel Shift) and would serve to indicate that the previous letters needed to be pronounced.

English is simply a web of stories like these. But unlike various other languages, it never got fixed.

French had some major sound-shifts, but they were fairly predictable consequences of the ways in which French articulates various sounds (things like a back "r" swallowing up following front consonants, or the combination front vowel + s + front consonant losing the "s," that sound being turned into the circumflex), and French had relatively little borrowing afterwards, due in no small part to the rising power of its own standardization bodies which were rigorously anti-borrowing. As a result, you can roughly guess how things are written. Spanish, likewise, had a major reform in the early 19th century.

Other languages were even more recent. Hebrew, for example, has had a notoriously Baroque spelling system, as it has what's technically called an "abjad" rather than an alphabet: that is, the letters represent consonants, and you just have to know the right vowels. As a result, the standard exam for prospective newscasters in Israel was to simply give them a sheet of text and have them read it out loud. Even fairly ordinary words would quickly start to read like Trenité's poem, and very few people could read it all correctly. The language had a major spelling cleanup in 1996, standardizing on the "ktav malé" system, and it's now possible for even people without profound mastery of the language to read an arbitrary piece of text out loud. 

So while English isn't the worst language to learn, by far, it has ended up with a complete mess of a spelling system.

It could be worse; it could be Chinese. 


You can find the full text of the poem at https://web.archive.org/web/20050415131319/http://www.spellingsociety.org/journals/j17/caos.php .
A good history of English spelling is at http://www.ruf.rice.edu/~kemmer/Histengl/spelling.html , and you can read about the Great Vowel Shift at http://en.wikipedia.org/wiki/Great_Vowel_Shift .
98 comments on original post
2
Add a comment...

Alex Borodach

Shared publicly  - 
 
"During the Vietnam War, to protest against American imperialism, he gave lectures on category theory in the forests around Hanoi while the city was being bombed"
A man with brightest mind, big heart and passion, RIP Alexander.
Very few people in our world can sacrifice everything for ideas they live for, not saying few people at all have ideas they live for.
http://www.telegraph.co.uk/news/obituaries/11231703/Alexander-Grothendieck-obituary.html
Alexander Grothendieck was a mathematician hailed as a genius who embraced 'militant activism’ before losing his reason
1
Add a comment...
 
New deep convolution neural network GoogLeNet can classify objects on images with 6.6% top-5 classification error over ImageNet image collection. Isn't it awesome? Certainly, not far those days when machines will outperform human in objects classification in general, a task considered one of the hardest in machine vision only 10 years ago.
 
Our GoogLeNet entry won the ImageNet 2014 detection and classification challenges!!!

Congratulations to my awesome teammates +Christian Szegedy, Wei Liu, +Yangqing Jia, Scott Reed, +Dragomir Anguelov+Dumitru Erhan and Andrew Rabinovich. Thanks to everyone who contributed to the Google Brain project as well!

After joining Google Research a few months ago, I gathered this crack team and drove the project. We obtained substantial improvements over last year's results (overall about twice better) with 6.6% top-5 classification error (almost 2x better than 11.7%) and 43.9% mAP detection accuracy (almost 2x better than 22.6%).
We did not use any external data except for the ILSVRC12 classification data for pre-training our detection model.

The improvements stem from a more sophisticated ConvNet architecture by Christian and the hard work of the entire team. We submitted to the "open" track, which means that we will be revealing the details of our system, so stay tuned.

The name of our submission is a tribute to +Yann LeCun's original LeNet convolutional networks.
Team name, Entry description, Number of object categories won, mean AP. NUS, Multiple Model Fusion with Context Rescoring, 106, 0.37212. MSRA Visual Computing, A combination of multiple SPP-net-based models (no outside data), 45, 0.351103. UvA-Euvision, Deep learning with provided data ...
17 comments on original post
1
Add a comment...
Have him in circles
182 people
peerawat sertpan's profile photo
Jadviha Sikorskaya's profile photo
Натали Горбунова's profile photo
Виктор Букато's profile photo
Влад Латынский's profile photo
Than joung phaoh's profile photo
Igor Karanevich's profile photo
MILOS JANATA's profile photo
Aleksandr Semchenko's profile photo

Alex Borodach

Shared publicly  - 
1
Add a comment...

Alex Borodach

Shared publicly  - 
 
Interesting how special and general relativities used to correct clocks on GPS satellites. It turns out they counteract leaving some residual effect. Look obvious after you got to know this.
 
A Matter of Time

can be used to pinpoint your location. Each of the satellites transmits a signal with its current position and time. A GPS receiver can then use signals from multiple satellites to determine where it is on the planet. GPS is perhaps most widely used in mobile phones. If you’ve ever had your phone give you directions, you’ve used GPS. In order to work properly, the clocks on a GPS satellite need to be extraordinarily precise. So precise that the effects of special and general relativity must be taken into account.

In special relativity, the motion of an object relative to you causes the apparent time of that object be slower. Since the GPS satellites are orbiting the Earth, their clocks seem slower to us. In general relativity, a clock in the presence of a gravitational field run a bit more slowly than one without. Since the gravity at the surface of the Earth is a bit stronger than at the GPS orbits, the clocks of the satellites seem a bit faster to us. When you combine these two effects you get that the GPS clocks run too fast by about 38 microseconds a day. In order to correctly position objects on Earth we need a precision of about 20 nanoseconds a day, which is more than 1,000 times more precise. The GPS clocks are adjusted to account for time dilation, so that they broadcast Earth-time, not their local time. So every time you use your phone to locate the nearest bookstore you are utilizing technology that absolutely depends on relativity.

Since the GPS clocks are so precise, any fluctuation in the gravity around Earth would affect their timing in a small way. This led to a new idea published in Nature Physics, that proposes using GPS to detect dark matter. As many of you know, dark matter is that mysterious type of matter that doesn’t interact strongly with light. There’s a lot of evidence to support its existence, ranging from gravitational lensing to the large scale structure of galaxy clusters, but we’ve never detected it directly, so we aren’t entirely sure what it is. There have been lots of proposed ideas, ranging from the ever popular WIMPs  to exotic particles such as axions, to primordial black holes.

This paper considers an even stranger dark matter candidate known as topological dark matter. Most dark matter consider the stuff to be some type of particle. These would interact gravitationally to clump on galactic and intergalactic scales, but not on smaller scales. With particle models there simply isn’t enough gravitational warping by dark matter to have a measurable effect. In fact, studies of stars within about 1,000 light years of Earth shows no net gravitational effect from dark matter, which (somewhat counterintuitively) is what we would expect.

Topological dark matter (also known as top down dark matter, or TD) is very different. Instead of being some kind of strange particle, TD matter is generated by topological defects in space and time. The idea is based upon the cosmic string model (not to be confused with string theory). To use a very rough analogy, as water freezes to ice, cracks can form at the intersection between two regions of the water that started to crystallize. In a similar way, during the inflationary period of the early universe, different sections of spacetime “freeze” into their current state, leaving cosmic strings like cracks in the fabric of spacetime. In the TD model, these cosmic strings can gradually decay to create various particles, including dark matter.

Because these topological defects are localized in space, newly created dark matter would tend to clump around it. Such a clump would distort space and time around it. Not much, but enough to be measured with precise clocks such as those in GPS. In this new paper, the authors look at how GPS clocks would be affected by a clump of TD dark matter that happens to pass near Earth. What they find is that it could have a measurable effect.

Just to be clear, there’s no evidence of cosmic strings and the like, so topological dark matter is a bit speculative. However given the number of dark matter candidates that have been eliminated, it is definitely worth checking out. What’s particularly clever about this idea is that it doesn’t require the construction of a new dark matter detector. The advanced lab is already orbiting Earth. All we need to do is analyze data that’s already there, which the authors have started doing.

So it’s possible that finding dark matter may simply be a matter of time.

Image: US National Executive Committee for Space-Based Positioning, Navigation and Timing

Paper: A. Derevianko & M. Pospelov. Hunting for topological dark matter with atomic clocks. Nature Physics]doi:10.1038/nphys3137 (2014)
17 comments on original post
1
1
Dzmitry Lazerka's profile photo
 
Cosmic strings sound very fishy :)
Add a comment...

Alex Borodach

Shared publicly  - 
1
Add a comment...
 
#Manhattan (August 2014)
1
Add a comment...
People
Have him in circles
182 people
peerawat sertpan's profile photo
Jadviha Sikorskaya's profile photo
Натали Горбунова's profile photo
Виктор Букато's profile photo
Влад Латынский's profile photo
Than joung phaoh's profile photo
Igor Karanevich's profile photo
MILOS JANATA's profile photo
Aleksandr Semchenko's profile photo
Education
  • Belarusian State University
    Ms Computer Science
Basic Information
Gender
Male
Other names
Alexander
Work
Occupation
Software Engineer
Skills
JavaScript
Employment
  • Startups
    2014 - present
  • Google
    2012 - 2014
  • Yandex
    2011 - 2012
  • Startups
    magician, 2012
  • EPAM Systems
  • ScienceSoft Inc
Places
Map of the places this user has livedMap of the places this user has livedMap of the places this user has lived
Currently
Minsk
Previously
Mountain View - Москва - Несвиж
Public - 2 years ago
reviewed 2 years ago
1 review
Map
Map
Map