Powered by glolg
Display Preferences Most Recent Entries Chatterbox Blog Links Site Statistics Category Tags About Me, Myself and Gilbert XML RSS Feed
Sunday, Apr 22, 2012 - 01:03 SGT
Posted By: Gilbert

Eye To Eye

After a particularly productive night of Meepo (his clones appear to have a HP bonus? still not my cup of tea), Skeleton King (need to remember that Reincarnate costs mana! Solid, but turns like a battleship), Doom Bringer (got HP, can farm, but armour poor - ranged Terrorblade bleedin' hurts), Rikimaru (the A.I. actually buys wards! too bad that doesn't help much against a Butterfly and three Eaglehorns) and finally, Lich (infinite mana = infinite frost nova = standing between opponent hero and the creeps... bliss!), I woke up relatively early to get a pre-LASIK evaluation.

A circuit of staring into various machines and pupil-dilating eyedrops later, I finally got to the (bespectacled) doctor, who informed me that the thickness of my cornea was unfortunately insufficient, and was perhaps prepared for rather more disappointment than actually transpired - as inborn attributes go, having thinner corneas (in the bottom [or top, as he kindly tried to spin it] 3% of the population) is certainly something that I can live with. He suggested implantable contact lenses if I was adamant, which I was not, and thus I left the clinic with my requested three colourful pages of data and another task crossed off my list.


Glorious, glorious data


This however still leaves me with the outstanding problem of flying glasses during sports (strangely, far less in badminton, which may have to do with there being far fewer involuntary sharp turns), which motivated the visit in the first place. It may be time to acquire some goggles. Oh, and Dota 2 has lived up to my expectations so far, even if the binding of skill keys to QWER is taking some getting used to.


Academia Review (Continued)

Proceeding from where we (here, the academic we) have left off, at citations and the h-index, there is another route beneficial to clocking them, and that is collaboration - gone are the days when most scientific treatises had single authors - three or more (which can go into the thousands!) is the norm nowadays. While the order traditionally has some significance, with first authorship somewhat prestigious (though not always, and sometimes exploited for humourous effect), this isn't reflected in most common metrics.

Quite clearly, then, having a good amount of collaboration is highly beneficial to such statistics. Imagine a laboratory of twelve moderately-productive research assistants and graduate students, whom can on their own write an average of two publishable papers per year. However, if they pool their creative efforts, and thereby qualify for authorship on each others' papers, they will then each be the proud author of 24 papers a year, a trememdous output! While this may be an exaggeration, it still illustrates how having a few buddies (to actually bounce ideas off and share in the work) can double or triple one's productivity with little additional effort.

This does mean that raw citation metrics alone may not be the best indicator of a researcher's output, and the obvious correction is to divide the credit for each paper between all the authors (so, for example, a paper with four authors and 100 citations counts as 25 citations for each in their individual stats, instead of 100 for all four). This might however create some awkwardness among the friendlier section of the academic community, and we wouldn't want that, would we? Constructing a PageRank-like graph for papers (where citations from more heavily-cited papers count for more) could also give interesting insights - and strategies?

Continuing on, what is considered novel enough to publish? Papers do have to be different (though reproduction of results doesn't get enough love in my opinion) from what has come before (if only not to be mere plagarism), but deciding what is novel is surprisingly complex, not merely "I know it when I see it"; there are the outliers, and then there are those discoveries that seem kind of obvious (as in reading the problem statement allows one to guess the approximate methodology - but then, there are generally only so many...)

It might be useful to consider a concrete example. Take for instance bilateral filtering, a very popular method introduced in 1998, which moreover has the property of being explainable to laymen in visual terms. Oftentimes, it is desirable to clean up an image by removing much of the noise, if only to save some of the millions of hours expended by girls attempting to painstakingly airbrush themselves into modelhood (but often into mannequinhood instead)

For a first attempt, we might try the Gaussian filter (which has been known since what can only be described as antiquity), or to put it simply, replace the value of each pixel in the image with a weighted average of the surrounding pixels. Simple enough, no?


(Original source: vision.ai.uiuc.edu)


The original image is at the left, with the middle and right faces having Gaussians of increasing strength applied; while possibly-unwanted facial texture is removed by the blurring, the results are not exactly acceptable. [N.B. There is nothing particularly special about the Gaussian kernel in this application - a simple box filter (an unweighted average) might do for some images, the Gaussian itself can be well approximated by a triangular filter (linked paper does the math lifting), and as it happens, by repeated application of a box filter, which may not be exactly the same but hey, it's all approximate (discrete) on a computer in the end]

Now, bilateral filtering. While the equations may look a teeny bit threatening to those who have been away from school for some time, the gist of it is that instead of averaging all pixels in a neighbourhood, only similar-enough pixels (parameters to be set) are averaged - not particularly unimaginable on hindsight (and indeed, a special case named G-neighbours had been devised some time ago, as was readily acknowledged by the authors)

[N.B. Another popular option is anisotropic diffusion, which may be slightly more involved and seems to have its roots in physics, like a number of newer fields, though it may not be that unrelated to bilteral filtering after all]

The results are understandably pleasing:


Strictly speaking, another approximation


And where can we go from here? The natural extension would be to trilateral filtering, which as it happens has been attempted by many authors, but has yet to really catch on, the probable issue being that finding a relevant "third dimension" that offers an appreciable improvement is not easy. It could of course proceed to nth-lateral filtering in time (visions of razors begin to dance in my mind), but that somehow feels a bit like general clustering/classification.

学也，不外拆、合、修、伸、释；聲不過五，五聲之變，不可勝聽也。色不過五，五色之變，不可勝觀也。味不過五，五味之變，不可勝嘗也。

Taking a wider view, there is no reason why bi/tri/nth-lateral filtering can't be applied to less-structured graphs (for an image is after all just a regular one) - but what does it do then? Skipping to another area, did the inventors of the Fibonacci heap aim to produce something with a better-than-binomial heap runtime from the bottom up, or did they just start from the definition? Probably either way could have worked, so long as they actually did something.

At the end, every (good) paper has to offer something, be it a previously unattainable effect, better results on some (hopefully public) dataset (the representativeness of datasets is itself a deep problem), a faster runtime, more robustness, ease of implementation, etc; of course, while in theory weaknesses in a method should be discussed, this is often cursory and left to future authors to point out and justify their own niche, especially in mature fields.

So what's good enough to publish? I suspect I don't quite know, having had essays I thought were excellent unappreciated and marked down, and what I considered run-of-the-mill dross marked up. It therefore appears that a good solution would just to work on a lot of stuff and hock them to good-enough conferences and journals, and if nothing get feedback - and who knows, the supposedly less-promising ones may even turn out the more useful - I have a suspicion that one's number of publications/citations is at least moderately related to the number of attempts, if that is what one is interested in. Can't score on shots you don't take!

[N.B. The definition of a good-enough outlet can be subjective, though tier/ranking lists are openly available online (despite being supposed to be private, at least in the past, to prevent hurt feelings) and tend to agree with each other. This ranking seems to be in large part based on acceptance rate (like universities, come to think of it, though again cause and effect is blurry) - 20% or less for some top-tiers, about 50% for most other respectable conferences, and so on down to the pay-up-and-get-in vanity press equivalents]

One final note - the writing of a paper itself is usually not my biggest concern, though squeezing down to a page limit is often harder than filling up empty space. Some knowledge of LaTeX can be very helpful, since it produces some of the most beautiful typesetting (inclusive of mathematical formulae) in the business, and is freely available.


sin(Flesh)n

There is news, and then there's news, and the dismantling of a vice ring, which would have been wholly unremarkable had not one of its members (whom court documents have unimaginatively referred to as Miss XXX) been (marginally) underaged - and prolific. This might still not caused that much of a commotion, had not the ringleader not maintained extremely detailed records of her exploits (for what, seriously? to give loyalty discounts?)


Mr. Paper Bag would like to assure readers that this man, contrary to unfounded rumours, is not him. Like he would ever need to pay for it.
(Source: thejakartapost.com)


Her clientele, as it turns out, reads like a who's who of the pillars of society, including but not limited to a secondary school principal, a high-flying scholar teacher, a police superintendent, a lawyer, an executive director, an investment banker and numerous businessmen besides, ranging in age from 21 to 48, and including several races and nationalities in the best spirit of Singaporean diversity and hospitality (as is usual, our guests are free to skip town - have a good day!). This has even made the rounds of the international news, which should help reassure a certain famous liberal arts university that we are actually human over here, hello.

Well, it should be said that the pillars of the pillars of society have needs too, and it is extremely heartening to witness such singleminded unity of purpose in these testing times for a wounded national psyche riven by various divisive concerns; I am confident that should their energies be put to more conventional use in time to come, the scourge of our dismal birth rate shall be eradicated for evermore*. Additionally, moral to the last, the educators took the opportunity to set a good example by announcing their intention to plead guilty.

Maybe this inclusiveness is not absolute - there is a disproportionate proportion of especially successful men in the list, which can probably be put down to her services costing up to S$750 a pop; I have no reason to suspect that lower-income male citizens were relatively more averse to her charms as a group, other than the matter of the price. Ironically, this limitation likely guided them to more traditional establishments down south, where they had their needs met quietly, safely and completely legally, without threat of recrimination and public shaming.

[*I am otherwise not particularly confident on this, especially given possibly unrectifiable hypergamous tendencies in a society where the gender distribution of educational qualifications are rapidly becoming equalized.


Basically it's like this
(Source: cutestuff.co)


When the national news came out with freezing eggs to reverse the falling birth rate, I concluded that they were finally hopelessly out of ideas and that the native population is doomed]

There has been a considerable amount of sympathy for the men involved, given that they could not be expected to know that the lady in question was too young. Furthermore, since time immemorial, it is a truth universally acknowledged that a (not necessarily single) man in possession of a good fortune must be in want of a nice willing lady to whom he can happily gift some of his excess wealth in exchange for a mutually beneficial healthy workout, crucial human contact, and perhaps, if performance is satisfactory, a long-term arrangement in which the lady is unionized as a "wife".

[N.B. And hey, even Gandhi, who has anchored so many admissions essays by candidates seeking a failsale role model, had a documented penchant for sleeping with young girls - bet that doesn't come out in the "Why I admire him" section often. It is astonishing how multi-faceted even widely-accepted avatars of virtue tend to be on closer examination, but well, the world needs its heroes...]

Therefore, it has to be hoped that the men learn their lesson and keep to more orthodox practices (as one of my JC's alumni has apparently figured out) - if they feel that patronizing common bordellos is beneath their station, they might be best served visiting places called "clubs", where they can purchase a steady stream of alcoholic beverages for the lady of their choice, and perhaps some flowers (though this has become rather institutionalized) and expensive trinkets (no direct exchange of cash, to be on the safe side, even if it introduces economic inefficiency into the picture), in return for a fair chance at getting lucky. If worst comes to worst again, an expensive lawyer can probably push some of the blame onto the establishment for not checking her identification.

However, how evil were they really? Lest I be strung up immediately, how many of the living are not born of flesh? Heck, even deities conventionally choose to enter the world thus! From this perspective, it is kind of odd that the act is so taboo, given that the vast majority does, or will do, it. True, it may not be tactful to discuss it in public (like bowel movements), but this does not detract from its utter necessity and ubiquity. Which reminds me, Mr. Ham has been going on all day about how if his idol the Dear Leader had let himself go to the toilet once in a while, he might yet be with us.

While probably not applicable in this case, as my former Chinese teacher once nobly expressed (paraphrased), "青楼女子，有些遭遇是很可怜的。", [N.B. the Green Ajah of the Tower coincidentally also... likes men] they were quite often led to their profession, honoured as the first and the oldest, and historically offering fair prospects for advancement, by circumstance. As there is little indication that it will ever disappear as long as men live, one might expect that they at least be kind and unfailingly polite throughout the transaction, unless the lady's tastes run otherwise, in which case certain accomodations might be made - as only befits the duty of a proper gentleman.


Mr. Ham (2847/2800 seeds): The underhams must come good eventually - Wolverhampton (-1.5) vs. Manchester City (at 2.15)

FAKEBERT (2945.5/2800 seeds): No value to be had, but Manchester City to beat Wolves (1.18) should be as sure as they come!



comments (0) - email - share - print - direct link
trackbacks (0) - trackback url


Next: Midweek Update On Data Cells


Related Posts:
Another World
Facts About Life
Fancy Footlose For Free
Holidays Over
Superfreakidocious
Back to top

Copyright © 2006-2025 GLYS. All Rights Reserved.