Powered by glolg
Display Preferences Most Recent Entries Chatterbox Blog Links Site Statistics Category Tags About Me, Myself and Gilbert XML RSS Feed
Saturday, May 05, 2012 - 19:12 SGT
Posted By: Gilbert

Fixy Fixed



Specs Savers

It's been a long time, but it's finally done:


Copper wire and bondo, outperforming cable ties since 1950


Oh, and the blind are being healed... by science.


The Search For Novelty

It may sound bad, but I honestly feel dumber as the graduate course winds on; apparently the realisation of what one doesn't know grows much faster than what one gets to know. I've got to think about how to get out before I regress into a gibbering idiot, which should at least raise my social stock.

The trouble is coming up with something sufficiently novel, as the more I read, the less I believe that anything new (at any rate, new and approachable) has ever existed. Oh, there's of course putting a twist on prior art, or squashing them together, such that they yield some slight improvement on some special case that will be argued to be more general than it really is - which is probably what I'll end up doing. More bashing pieces into new configurations, then (though it may be one of the last to be discovered without brute force...)

On research, fashions do come and go, as one Google employee noted back in 2005, even before when I toyed about with Google Trends (yes, it's still around). I figured I could do a quick update using Google Ngram Viewer, which puts the statistics of all those books they aggressively scanned (sometimes without asking, but hey, they are data incarnate) at everybody's fingertips. Note that the occurrences are normalised by the quantity of books published each year:



As is confirmed, while references to "machine learning" are steadily (if slowly) growing, "expert systems" fell off a cliff towards the end of the 1980s, probably about when people began to realise that they were basically huge and unwieldy decision trees. Neural networks too hit a hump towards the end of the 1990s, to the point that one of its founders observed wryly that new researchers who came in after 2000 may not have heard of it (it does probably draws its base from further afield than just computer science), while genetic algorithms have kind of plateaued.

[Notably, there exists a strange hump for these disciplines between 1900 and 1910; either someone was seriously ahead of his time, or some publishers felt it safe to omit the first two digits of the year. Then again, Mr. Robo assures me that all these are basically interchangeable, as they should all be transformable to and from, say, a (possibly complicated) decision tree. I'll take his word for it.]




The increase in mentions of support vector machines have not shown any sign of slowing... yet (though they are still a blip in absolute terms), the same for random forests, to a smaller extent. There is, certainly, debate as to which of these is best for a certain domain (of especial interest to those who actually need concrete results), but this will probably never be fully answered, in the sense that whether some selected subset is representative of a domain is probably already impossible to prove; it is true, though, that a simple method used properly may well be better than a sophisticated one used poorly (SVM and parameters, I'm looking at you)

[To be continued with an applicative example... soon.]


The House Debates


In the M'sian Parliament recently, an UMNO MP told a story:

"There was a father who gave RM100 to each of his 3 sons and asked them to buy things and fill up a room completely. The first son bought hay for RM100, but couldn't fill the room entirely. The second son bought cotton for RM100 but couldn't fill it either. Finally, the third son bought a candle for RM1, lit it up, and the room was filled with light!"

The proud UMNO MP then declared:
"Our Prime Minister is like the third son.
From the day he has taken charge of his office,
our country is filled with the bright light of prosperity!"

Then, a faint voice from the backbenches asked:
"But where is the remaining RM99?"



Me: Well, well, they haven't been too bad of late, coming to an agreement to allow private jetties at Seletar, pending their refurbishment; it's all these little things that stick in the mind. And the union fellows have finally remembered their job scope, even if it's just talk. And a think-tank worth the name has realised the basic logic that pumping people in (while shouting down dissenters with accusations of xenophobia) without fixing the birth rate is equivalent to kicking a grenade down the road. Things may finally be looking up!

Mr. Ham: Seeing as to what being here does to the urge to procreate, importing cadavers might have a more desirable demographic outcome in the long term.

Me: Don't be such a hater, they even made a profit from selling stock in Chinese banks! Granted, part of it can be attributed to being a sweetheart pre-IPO investor, but I'm glad we got the better of the market for once.

Mr. Ham: You might want to reconsider the "we" part, though in retrospect even they couldn't go wrong with CCB, which as we have seen maintains undiminished demand whatever its condition and price. For a small fee, I will have Esquire Pants write in to the papers to claim proper credit and recompense for the investment strategy employed.

Me: No need lah, no need lah.

Mr. Ham: Continuing on, there may be something in the water at the Ministry of National Development - why else would they claim that HDB flat sizes have not been shrinking, when they candidly admitted to it six months ago? They might consider hiring a consultant dedicated to keeping the spin straight, and I know just the ham.

...and why am I saying all the bad things today?

Me: I don't want to lim kopi mah.

Mr. Ham: Me neither, I much prefer tea... oh. How about let's talk about a principality called Hamopolis instead?

Me: Hamopolis, sure.

Mr. Ham: Now this place Hamopolis, its public transport system is falling apart, very coincidentally like here, though there is otherwise completely no connection between them, of course. There have been calls to nationalize it in response...

Me: Like here, again?

Mr. Ham: ...but the catch is, in Hamopolis, the citizenry doesn't realise that their system is already nationalized!

Me: Huh?

Mr. Ham: Let me explain with some simple infographics. This, I believe, is what most people have in mind when they think of "nationalization":



In this model, the relevent government ministry builds and administers the system, which likely requires a huge (but justified) initial investment, which happily for Hamopolis is not a problem due to their mighty reserves (which came from taxes in the first place). In an ideal case, the system is funded and maintained by a mixture of fares (and taxes if necessary) such that it runs without a deficit or surplus in equilibrium. The people in charge do their jobs and draw reasonable wages for the not overly taxing task of administering a monopoly of an extremely inelastic good.

Me: But won't there be inefficiencies without incentives, leading to oversights like overpaying on quality parts or maintenance?

Mr. Ham: About that, observe the Hamopolis model:



Marvel in the beauty of the setup: Ministry T, which sounds like it should be in charge, has next to no say in practice, but has the job of appealing to Ministry F for financing when things go pear-shaped, which Ministry F has to approve after grumbling since, like big banks, public transport is too big to fail; but that's not a problem, Ministry F controls an external investment company which runs the system in the first place, and that brave paragon of competitive capitalism rakes in huge profits, year on year, which it then invests very wisely.

Simplified further by removing some unimportant links in the chain and focusing on the inputs and outputs:



Me: Shouldn't the people of Hamopolis be mildly annoyed?

Mr. Ham: Well, they do have the consolation of an extremely successful investment company that they can be proud of. And there's more, they did it without touching any pension funds:




So, as you can see, the investment company definitely does not manage any of the pension money. The money went to the board, which did the responsible thing and bought bonds from one of the most stable governments around. All this money was then locked in one vault, while some other money was handed by the government to the investment company. No connection at all!

Me: Um, that isn't too convincing...

Mr. Ham: I thought you had a degree in economics? This is basic stuff I learnt in my first year at Thug U - after you get funds from some source, like loansharking, all you have to do is to pump it through some legitimate business, like the convenient neighbourhood integrated resort or laundromart, and there is then absolutely no link between what sits in your bank account and the original act!

Me: Still sounds risky to me. Thankfully, that could never happen here, since a minister helpfully dumbed down the relationship to us and clarified that no CPF funds go to our government-linked investment companies.

Mr. Ham: You know, a China-based academic recently published a research note accusing one of those companies of achieving unrealistic Buffettesque annual returns of 17% over 30-odd years, in view of the fact that most of their investments have been domestic until recently, and the domestic market return has been far from that. Evidently he hasn't taken into account the value of all the former state assets that just happen to fall into their lap every now and then as and when required; all that included, I would be disappointed if they didn't hit 17%!


Flip Flops

Me: Whatever lah, I think we can still trust the government. *glances around*

Mr. Ham: Not like you have much of a choice there. From my perspective, them trying... things may not automatically be bad - but they had better be an Ashley Young going about it, instead of being Andy Carroll:


...can't even give one point for that
(Source: whoateallthepies.tv)


Me: I hope, I hope. For FAKEBERT (3198.5/3000 seeds), Newcastle (+1.5) to stick it out against City (at 1.50)

Mr. Ham (3027/3000 seeds): Making it easy, Fulham to beat Sunderland (at 1.80)



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


Next: Fighting Words


Related Posts:
Economics Thus Far
Open Bookame
On Economics
STATE FAIL
Taking Stock
Back to top

Copyright © 2006-2025 GLYS. All Rights Reserved.