Tuesday, January 26, 2016

How much time should I give myself to catch a connecting flight at Changi Airport without baggage?

Arrival

Just like what Eric has said, the worst case is if your AirAsia flight arrives at the furthest gate from immigration. Unfortunately, if I fly with AirAsia or Jetstar most certainly I will depart/arrive from far gates (C16-19, C24-26, D35-38, D46-49). In addition to that, when you arrive to Singapore sometimes there will be a random check on handcarry item (I encountered it a few times, and perhaps due to Jakarta bombing last week there will be checking again).Walking takes around 10 minutes, x-ray check takes at most 5 minutes, immigration takes around 10-15 minutes. So you will spend around 30 minutes for your arrival.


Transfer

This depends on your luck. You will have to take Skytrain from T1 to T3 (go up to departure hall first). If you arrive from C concourse, you will be lucky because C is nearby Skytrain to T3; but if you arrive from D concourse (which is close to T2) you have to walk quite longer. Worst case will take you around 10 minutes until you reach Terminal 3.
When you arrive at Terminal 3, the closest check-in row is row 11 while Lion Air is located at.... row 1. So you have to walk to the other end of departure hall. Will take you around 3-5 minutes and in total to change terminal you need around 15 minutes.

Departure

I've flown with JT163 (Lion Air's Singapore - Jakarta afternoon flight) for several times. Checking-in your flight will not be very long as Changi is very efficient, so I think 5-10 minutes must be enough for check-in. The bad thing is about gate. The flight is often (if not always) placed at A16-A20 gates, and walking from departure immigration to the gate can take 15 minutes alone. So far (surprisingly) I have never experienced any delay with this flight, so if you often hear about Lion Air's bad reputation in terms of on-time performance I don't think you should expect that in this flight. I guess for departure you need 30 minutes.

Of course my approximation is a bit exaggerated, but it is better to spare some time right? So I recommend you to take the earlier flight. Good luck, and I hope your trip will go smoothly :)

What is an intuitive explanation of the natural proof, relativization, and algebrization barriers to solving the PP vs. NPNP problem?

"Intuitive" is hard because some of these results are pretty nonintuitive.  I'll just tackle the relativization barrier.
The Baker-Gill-Solovay theorem is that there is some oracle A "relative to which" PA=NPAPA=NPA and a different oracle B, for which PB≠NPBPB≠NPB.  What does this formalism mean?  Normally when we talk about the complexity class PP it refers to languages recognized by Turing machines in a polynomially-bounded amount of time..  But for PAPA, the Turing machine has a subroutine (the oracle) which can recognize any language AA, or a class of languages AA, in one step.

For example, P3SATP3SAT is the complexity class of languages which a Turing machine can recognize in polynomial time, if it has access to a magic 3SAT solver.  This complexity class is at least as large as NP, since 3SAT is NP-complete.    PPPP is an example of using an entire class as the oracle, rather than an individual language.
What the theorem showed is that an oracle for TQBF (True Quantified Boolean Formulas) is strong enough to make PTQBF=NPTQBFPTQBF=NPTQBF.  TQBF instances are logical formulae containing existential and universal quantifiers--- that is, statements like "for all X, there exists Y and Z, such that X and Y imply not-Z".  TQBF is a PSPACE-complete problem.
The details are technical but basically the proof looks like NPTQBF⊆NPSPACETQBF⊆NPSPACE=PSPACE⊆PTQBFNPTQBF⊆NPSPACETQBF⊆NPSPACE=PSPACE⊆PTQBF.   TQBF doesn't add anything to NPSPACE (nondeterministic polynomial space); a previous theorem shows NPSPACE = PSPACE; but because TQBF is NPSPACE-complete, a Turing machine with access to an TQBF oracle must be at least as strong as PSPACE.
The other half of the theorem is coming up with an oracle B that definitively makes nondeterminism stronger, so that PB≠NPBPB≠NPB.  The general approach is to define a very simple language: SL(B)SL(B) which recognizes all strings exactly the same length as something in language B!  (Which we haven't defined yet.)  Obviously an nondeterminstic Turing machine can just "guess" some word of the right length and then use the B-oracle to verify that the word is in B.  The hard part of the proof is coming up with some B that a deterministic Turing machine can't generate the inputs to, in polynomial time--- so the oracle doesn't help.  The proof uses diagonalization, but the details are really hairy.
OK, so what does that show?
What the theorem gives us is that any "proof" or proof technique for P=NPP=NP or P≠NPP≠NP that isn't sensitive to the presence of an oracle, cannot work.  That's because we showed that, depending on the oracle, the two could be equal or not equal.
Suppose a colleague slips us a paper purporting to show P≠NPP≠NP.  We can search-and-replace every occurrence of PP by PTQBFPTQBF and NPNP by NPTQBFNPTQBF.  Now obviously the proof must now be flawed, because the substituted classes are equal.  So (if the proof is correct) somewhere it must make use of a property of PP that is not true of PTQBFPTQBF.
"Duh!" you might say.  Well, there is one very common proof technique which relativizes (i.e it is not sensitive to the presence of an oracle): Diagonalization.
The Time Hierarchy Theorem was proven using diagonalization.  It says, roughly, that for every big-O time class, there are problems that take at least that much time on a Turing machine.  So there are problems that take O(n5000)O(n5000) time to solve but can't be solved in O(n4999)O(n4999).  The proof is equally valid if we substitute in a Turing machine with an oracle!  Even our monster PTQBFPTQBF has problems that can be solved in O(n2)O(n2) but not in O(n)O(n).
Simple diagonalization is essentially a "counting" argument, which doesn't depend on the structure of the problems involved, only the final result.  And thus it can't solve P vs. NP.  (More sophisticated versions of the proof technique exist, though.)
An analogy that may be helpful is proving statements about numbers.  There are some statements you can prove about integers that are independent of whether you're working in ZZ or in ZmodpZmodp.  For example, we can show that 2+2=2∗22+2=2∗2 in any number system obeying the usual definitions of addition and multiplication.  But in some systems 2+3≠02+3≠0 and in other systems 2+3=02+3=0.  Thus, our number theory proofs "relativize" if they are ignorant of the "mod p" but are "nonrelativizing" if they depend on the particular modulus, or lack thereof.

What are some things that happen in movies that most people think are bullshit, but are actually true?

I don't know if this is something people would "call bullshit" on, but it's a pretty cool moment captured on camera.

In Quentin Tarantino's Django Unchained, Calvin Candie (Leonardo DiCaprio) confronts Django (Jamie Foxx) and Dr. Schultz (Christoph Waltz) in arguably the most climactic scene of the entire movie. In a fit of rage, Candie slams his hand down on the table and accidentally crushes a wine glass, leaving a large gash in his hand. Without breaking character, DiCaprio finishes the scene with copious amounts of his blood dripping down his hand and arm. He even rubs the blood on the face of Broomhilda von Shaft (Kerry Washington) to her horror, quite understandably.

After reviewing the takes, the one mentioned above was selected for the final cut (no pun intended). He later received stitches for the wound and mentioned it during several interviews.

Whenever I watch the movie, I wait for that scene in anticipation and am awed by the dedication of DiCaprio to his character as well as the reactions from the rest of that incredibly talented ensemble.

Why does Germany still not have veto power in the UN, considering they are one of the world's leading economies?

The UN security council permanent members are composed of the major allies involved in WWII victory, i.e. the United States of America, the Soviet Union (replaced by the Russian Federation), the Republic of China (replaced by the People's Republic of China), France, and the United Kingdom. As Germany was part of the Axis, they did not have a UN security council permanent seat when the UN was formed, and there are no provisions in the United Nations Charter for changing the UN's structure.

And while it could make sense to give Germany a permanent seat, it would face much opposition as well:

  • Germanophobia is in full swing in Europe even today as people from the PIGS (Portugal, Ireland, Greece, Spain) resent Germany's strong role in preserving the European Union. One wonders just how much it will be widespread if Germany assumes a bigger international presence.
  • Germans themselves may not actually want a bigger international presence.


It also opens a pandora's box of which other countries should have a permanent seat.

  • India did not exist as a country at the time of the UN's founding, and should have a seat for all kinds of reasons (economic, demographic, geographic)
  • There is no African country with a permanent seat.
  • There is no Latin American country with a permanent seat.
  • Should France and the UK still have a permanent seat, as there are already 3 European countries with a permanent seat?
  • There is no Muslim country with a permanent seat, should there be one? Turkey and Iran are likely the best candidate on paper, but are unlikely to ever be accepted by the World or by Muslims themselves. Indonesia could also be a good candidate on paper but it might also not be accepted.


I would personally support a German permanent membership of the UN security council, but getting this done is fraught with problems.

How far can artificial intelligence go?

I. J. Good and Vernor Vinge noted that if humans could produce smarter than human intelligence, then so could it, only faster. Good called this phenomena an intelligence explosion. Vinge called it a singularity. Ray Kurzweil extends Moore's Law to project that global computing capacity will exceed the capacity of all human brains (at several petaflops and one petabyte per person) in the mid 2040's. He believes that a singularity will follow shortly afterward. This assumes that global computing capacity (operations per second, memory, and network bandwidth) continues to double every 1.5 years, as it has been doing since the early 20'th century.

Current global computing capacity is about 10^19 operations per second (OPS) and 10^22 bits of memory, assuming several billion computers and phones. In 30 years, these should increase by 6 orders of magnitude. Ten billion human brain sized neural networks with 10^14 connections each at a few bytes per connection, running at 100 Hz, would require roughly 10^26 OPS and 10^26 bits.
It is hard to predict what will happen next because our brains are not powerful enough to comprehend a vastly superior intelligence. Various people have predicted a virtual paradise with magic genies, or a robot apocalypse, or advanced civilization spreading across the galaxy, or a gray goo accident of self replicating nanobots. Vinge called the singularity an event horizon on the future. We could no more comprehend a godlike intelligence than the bacteria in our gut can comprehend human civilization.
Nevertheless, physics (as currently understood) places limits on the computing capacity of the universe. Flipping a qubit in time t requires energy h/2t, where h is Planck's constant, 6.626 x 10^-34 Joule seconds. Seth Lloyd, in Computational capacity of the universe, estimates that if all of the mass of the universe (about 10^53 Kg) were converted to 10^70 J of energy, it would be enough to perform about 10^120 qubit flip operations since the big bang 4 x 10^17 seconds ago (13.8 billion years). This value roughly agrees with the Bekenstein bound of the Hubble radius, which sets an upper bound on the entropy of the observable universe of 2.95 x 10^122 bits.
Writing a bit of memory, unlike flipping a qubit, is a statistically irreversible operation, which requires free energy kT ln 2, where T is the temperature and k is Boltzmann's constant, 1.38 x 10^-23 J/K. Taking T to be the cosmic microwave background temperature of 3 K, the most we could store using 10^70 J is about 10^92 bits. This roughly agrees with Lloyd's estimate of 10^90 bits, which he calculated by estimating the number of possible quantum states of all 10^80 atoms in the universe.
If we restrict our AI to the solar system and captured all of the sun's output of 3.8 x 10^26 W using a Dyson sphere with radius 10,000 AU and temperature 4 K, then we could perform 10^48 OPS (bit writes per second). To put this number in perspective, the evolution of human civilization from dirt 3.5 billion years ago required 10^48 DNA base copy operations and 10^50 RNA and amino acid transcription operations on 10^37 DNA bases over the last 10^17 seconds. Thus, our computer could simulate the evolution of humanity at the molecular level in a few minutes, a speedup of 10^15. Anything faster would require interstellar travel or speeding up the sun's energy output, perhaps by dropping a black hole into it. (A naive extrapolation of Moore's Law suggests this will happen in the year 2160, 75 years after we surpass the computing power of the biosphere.)
Note: to estimate the computational power of evolution, I am assuming 5 x 10^30 bacteria with a few million DNA bases each, and a similar amount of DNA in other organisms. I am assuming a replication time of 10^6 seconds per cell, and that DNA replication makes up 1% of cell metabolism. See also An Estimate of the Total DNA in the Biosphere.

Why is it when there is a humanitarian crisis like Ebola, it's America and the West who send aid, we don't hear of China, India or other BRICs countries sending aid yet they have just as much to lose if the virus gets out of hand?

You are wrong. In the Ebola fight, most of the western nations are way behind.

Cuba has been the leading country on the ground in the past several month. For an island nation of 11 million people, it has 471 physicians on the ground. France (through médecins sans frontières) has 250 physicians on the ground. China has over 200.

According to IMC, the U.S. has less than 10 doctors registered to volunteer to fight Ebola. Cuba leads fight against Ebola in Africa as west frets about border security.

The US has pledged $400 million in aid, which is great. But right now, there are literally hundreds of millions of dollars sitting there and WHO can't find people to use them. At the end of the day, you need people to treat the disease, and it's hard to recruit volunteers to deal with a disease with 50% fatality rate.

If you do some research in this area, you will see that Cuba is by far the leading medical-care contributor in the world. They often drop a couple thousand doctors to disaster areas to help treat the patients and train the local doctors. It's their tradition and something the Cubans really believe in. independent.co.ukCuban medics in Haiti put the world to shame

What is a 1up from microsoft paint, but still free & simple?

I second John Colagioia's view -- GIMP is free but not simple. It's very powerful, but awkward to learn, and unless something has changed it starts up incredibly slowly. I also recommend Paint.NET, which is not available at http:// paint dot net, but is available at getpaint.net. It's a good step up from Microsoft Paint.

What best explains why the United States intervened militarily in former Yugoslavia and not in Rwanda in the 1990s?

Cold war strategies, of course. Yugoslavia was an independent state, with no strong Soviet Union ties, nor connections with the West.

As long as the Soviet Union was in place, the war in Yugoslavia would not have occured as the Soviets would have backed the Serbs - their ally up until the rise of Tito's Yugoslavia  - and the situation would've gotten far too risky.

But the Soviet Union DID collapse in the early 90's and NATO knew that the newly formed Russia didn't have the power to react.

Thus, NATO countries (most prominently Germany, of which Croatia was an ally in WW II and beloved country for vacations) began to support and fuel Croatia's and Slovenia's rise to independence, so that they could become under NATO's sphere of influence.

Serbians in Croatia opposed the idea of an independent Croatia, since they didn't want to lose their nationality. They, instead, wanted the areas in which they lived to become Serbian. As for Slovenia, not many Serbs lived here so there was no war, although the Yugoslavian army did try to invade Slovenia, but was stopped in its tracks.

Now, the NATO planned on "seizing" Croatia and Slovenia, since Croatia has a widely stretching coastline, both interesting strategically and for tourists, and Slovenia was the most developed state of Yugoslavia, plus an easy target (because of its homogenity).

HOWEVER, Bosnia, of which the majority of citizens are muslim Bosniaks, but a population of 30% are Bosnian Serbs, and a remainder of 15% Bosnian Croats, was still left. The Bosniaks either didn't want to live in the remains of a Yugoslavia dominated by Serbs, or were attracted to the promises that NATO made to Croatia and Slovenia (maybe a combination of both). The problem is that NATO either completely forgot about Bosnia, or simply let the Bosniaks drive into war with Serbia, knowing that it would get crushed by the latter. This would of course cause for bad publicity for the Serbs, and a stronger case for Croatia who at this time lost several areas of its territory to Croat Serbs.

The rest is history. Bosnia called for independence, which the Bosnian Serbs AND Bosnian Croats neglected, after which both parties started attacking the Bosniaks - it's just that you'll never hear about the Bosnian Croats in Western media. Moreover, Bosniaks under the knowledge of NATO and the US allowed thousands of Mujahideen from the Middle East(the same rebels that fight for Al-Qaeda now) to come to Bosnia and fight against the Serbs.

All parties committed atrocities, of which Serbs the most. However, in the media only Serbs were accused, a similar thing of what happened in Syria with Assad. The difference is that now we know the rebels real face, through the internet. If it were not for those ISIS videos, I believe that we would still be believing that the latter is "a friendly rebellion fighting for a good cause".

ANYWAY, the horrible massacre of Srebrenica happened, which Serbs claim to be a killing of terrorist Mujahideen (since only men were killed) that operated in Serbian villages around Srebrenica for several years, killing and decapitating victims.

The Srebrenica massacre and other atrocities committed by the Serbs in term "justified" NATO intervention in Croatia, which caused an estimated 200.000 Croatian Serbs to flee. Most of their houses were burnt down and to this day they have still not returned.

After this, peace talks ended the war, and Yugoslavia was no more, all its former states except for Serbia now under NATO sphere of influence. To this day all of the former states, except for Slovenia, are poorer than they were before the war.

All sides committed horrible atrocities, of which Serbs the most, but only the Serbs got convicted.Tens of Serbian officers have been charged to a life in prison, while Bosnian rebel leader Naser Oric and Croatian general Ante Gotovina, who was sentenced to 30 years, but was released due to lack of evidence, still walk around freely.

I'm going to Siem Reap, Cambodia. Should I book tourist activities before I go or when I get there.?

We have a lower medium type budget. Also, is it worth the extra money to hire an English speaking tour guide? What are the average tipping prices there? Can you bargain with tourist services. My husband is Chinese and he isn't used to paying for things at the same American tourist rate.

Yoshua Bengio: What is the most exciting machine learning research paper you read in 2015?

No single paper stands out, and I realize talking to people that different researchers are impressed by different contributions, so the choice of the advances below is very subjective:

* the Batch Normalization paper is exciting because of the impact it already had in training numerous architectures, and it has been adopted as a standard

* the Ladder Networks paper is exciting because it is bringing back unsupervised learning ideas (here some particularly interesting stack of denoising autoencoders) into the competition with straight supervised learning, especially in a semi-supervised context

* this year's papers on generative adversarial networks (GAN), the LAPGAN and DCGAN, have really raised the bar on generative modelling of images in impressive ways, suddenly making this approach the leader and contributing to the spirit of rapid progress in unsupervised learning over the last year; they compete with another big advance in deep generative modelling based on variational autoencoders, including the very impressive DRAW paper from early last year.

* the papers that use content-based attention mechanisms have been numerous over the past year; I saw it start with our neural machine translation with attention, followed by the neural Turing machine (and later the end-to-end memory networks), and many exciting uses of this type of processing for things like caption generation and manipulating data structures (I liked in particular the Pointer Networks and the other papers on differentiable data structure operations with stacks, queues, Teaching Machines to Read and Comprehend, etc.). So this architectural device is here to stay...

Was Volkswagen actually the brainchild of Adolf Hitler, or is that a modern myth?

That's actually true. He wanted a car produced that had 4 seats, could hold a traveling speed of 100 km/h, was economical in its fuel needs and should cost no more than 1000 Reichsmark. Since the existing car companies did not think that this could be done, he tasked Ferdinand Porsche to design such a car and the Deutsche Arbeiterfront (German Workers Front) to build a factory to produce the car.

The result was the KdF-Wagen (KdF="Kraft durch Freude"="Power through joy"), a predecessor of the Volkswagen Beetle.
Fun fact: Wolfsburg, the city the Volkswagen company has its headquarters in, did not exist at that time. It developed around the new car factory and was first called "Stadt des Kdf-Wagens nahe Fallersleben" (City of the KdF-Wagen near Fallersleben).
There were plans to rename the city to Adolf-Hitler-Stadt (Adolf Hitler City), but this did not take place due to WW II. After the war the city was renamed to the name of a medieval castle nearby, called Wolfsburg (Wolf's Stronghold).

What is it like to work for the United Nations?

First of all, it was quite an honor. I was very young and had recently gotten out of the US Army and no US government agency wanted to hire me untiil after they'd seen that I'd qualified myself by getting hired at the UN first. Then the offers came but I rejected them because I was already at the UN. =)

Seriously, the vote of confidence I got from the premier international organization after learning German and Russian in the US armed forces was much better than the vote of zero confidence I got from US government agencies after having applied to them as well (they hadn't sent rejection letters; they just hadn't bothered to show interest).

The UN give a morning aptitude exam near the Manhattan compound to all those aspiring to work there. Rather than try to commute to midtown from my parents place in the suburbs (Rockland County), I arrived the day before the exam (which was given at least once a week) and stayed at a really cheap run-down hotel in Hell's Kitchen...before the rejuvenation of that area. I was determined to try to make it on my own and had refused to let my parents pay for a decent hotel. I still remember the dried blood on the hotel room walls that had been sprayed by heroin addicts; this signified the financial low point of my life in the moments before I took that UN exam.

I believe the results of the exam were immediate (my memory may have forgotten if there was an extra day's wait) and I went to the main recruiting office to formally present myself for a job with the test results in my hand.

Waiting for the job interview, I remember sitting with a recent American citizen graduate of John Hopkins who had majored in International Relations, undergrad or grad, I forget. As we talked, it was clear that he felt himself to be far more qualified than I was because he had such a low opinion of those who would serve in the US military. To him, it didn't matter that the Army had taught me Russian and German, fluently and that I'd served all over Europe and been all over the US and Hawaii. He assumed he was from a higher socio-economic class (even though I also had an undergraduate degree) and that he would get any job that was currently open.

He went into his interview first. When he came out 15 minutes later, he was ashen-faced and mortified. He had studied so many years at John Hopkins only to be told to go jump in the lake by the interviewer. He looked at me and said "Listen, if I couldn't get in, you might as well not even bother interviewing. There are no jobs available". Then he limped out as if his life were over.

That spooked me. When I heard my name called for my interview, I walked into the office, handed over my papers and said to the woman "I just heard from the John Hopkins grad that there are no jobs available, but I want you to know that, with my languages and military background, I'm highlly qualified to be a guard at the gates and will gladly accept $2K per month to be one."

She looked at me and laughed. She said "You mean that arrogant little fool that thought he was God's gift to the UN?" Of course there are jobs open. We just tell people whom we don't like that there aren't any."

Then she said "You've got the highest aptitude score I've ever seen. Now you could work as one of our brave and honored guards but I was thinking you might like a choice position that just opened as the archives manager for the Security Council Division on the 35th Floor of the Secretariat [ which of course is the tall building that symbolizes the organization at the corner of 1st Avenue and 42nd Street]. Let me make a phone call. Hold on." She then made a call and spoke in French to a woman who said I should come right over to be interviewed. I went over to the Secretariat and took an elevator to the 35th floor. The interview there went extremely well and my about-to-be-new-boss made a quick call to the Russian deputy to the Undersecretary General of that department. He was busy and said he trusted her opinion of the new potential hire and so I was hired. In retrospect, if this man had interviewed me and understood I'd just served with the US Army, he'd probably have said no out of fear that I was a CIA plant, which in fact I was not. The CIA hadn't been competent enough to even keep an eye on people getting out of the Army with skills they might want to exploit.

As it happened, I got along extremely well with all my Russian colleagues at the UN and they were glad, in retrospect, that I'd been hired even though the initial reaction on the 35th floor to having an ex-US miltary colleague was that of shock and acute discomfort (because they basically ran the 35th floor). I made good friendships with them and with political appointees from some very interesting countries and every moment at the Secretariat was fascinating.

Work started at 09:00 and ended at 17:00 for almost everyone at the Secretariat. The food and company in the cafeteria was fantastic at lunch-time and the place was packed. The view was of the East River right outside the glass wall and I remember the big neon Pepsi sign on the building in Queens across the river. It was used when filming "The Highlander".

Even better food was to be had when a top official invited me to eat in the ambassadorial dining room on an upper floor. That was posh.

The main cafeteria was open until 19:00 but 99.9% of UN employees refused to not go home when the bell rang at 17:00. There was never anyone to eat with in the vast dark cafeteria in the evenings. I never understood this because the food was great and low-cost. Why not eat before going home after work?

When the book "The Satanic Verses" was under attack at the UN, I played a not-insignificant role by accident. Just about every ambassador who was asked by their home government to condemn the book, was not allowed to read it himself. They all knew that their speeches were going to appear "uninformed". But, through the grapevine, a number of them learned that I had bought the book for myself at lunchtime the day before in a store on Second Avenue and that I had read the entire book overnight. So I had a parade of ambassadors coming to my office one afternoon (one after another) where we stood behind the filing cabinets and I pointed out the passages that I believed were rather controversial. They composed their speeches around my opinion of what was shocking.

Ironically, while I'm really into free speech, I'm proud that I was able to contribute to an informed discussion of the matter. Those speeches were accurate as to what Salmon Rushdie had written.

I also got on television receiving letters from protestors outside the buiding.

Sometimes in the late afternoon I'd be invited to parties being hosted by various embassies. The most interesting of these was that of Saddam Hussein's Iraq. While socializing, I witnessed an American who was obviously with the CIA who kept passing his business card to Iraqi embassy staff members. Following him was an eagle eyed Iraqi secret service member who made sure he re-collected every last card that had been given out. You could cut the tension with a knife there. I still had fun. I was there for the food.

I organized a big Christmas party that was canceled at the last second because of a plane crash. When UN employees are killed, everyone feels it. =(

By the way, the average UN secretariat employee was a woman from the Philippines because they were considered the least ideologically oriented and, thus, the most likely to be loyal to the UN itself.  

Now here's the kicker: The biggest perquisite for working at the UN was that I got invited to black tie balls on Park Avenue by people who were not related to the UN but respected people like me who worked there, regardless of their politcal opinion on whether taxpayer money was being wasted there. I'd be the first to admit that there were a lot of slackers sleeping on the job at the Secretariat. It still carried a ton of cachet to have a job there. The party invitations never stopped.

If my goal was to marry an heiress and be rich without having to work the rest of my life, I could have easily pulled that off in the year I was at the UN, because of all the cocktail parties and black tie parties. I could have quit the UN to be a Park Avenue house-husband or manage a wife's money.

But, one year later, I accepted a scholarship to take an MBA at the University of Toronto. I've been in business since then.

Part of the decision to leave the UN was that there was no real career path upwards because the top jobs, the boss spots such as ambassador or undersecretary, were all given by governments to political appointees assigned to the UN. If I want to be the US Ambassador to the UN, I will have to be personal friends with a future US President (which is entirely possible despite my reluctance so far to publicly discuss any political views in writing).

How to learn Machine Learning as a medical student?

It really depends on your background and how much mathematics can you understand. I will give a list of steps that you need to do to understand machine learning. Any comments are welcome to improve my answer as I am not the best machine learning scientist out there so far and I will improve my answer in the future. Please read the disclaimer in the end.

I assume that you have a basic understanding of mathematics to a college level*. I am also going to focus on the most basic things that you need to know; otherwise the background material might be overwhelming for you. I believe that learning comes from the experience, therefore you need to start with machine learning applications as soon as possible and learn the mathematics that you do not have studies on the first pass later.

First you need the mathematical background and you need to study at least the basics from the following topics.

Linear algebra.

  • You need to understand matrices and vectors, inverse matrices and matrix derivatives (see number 2).
  • You can study this topic from various sources and it is fairly easy. I have seen the following referenced:


  1. Linear algebra
  2. Linear Algebra Done Right, Sheldon Axler
  3. The Matrix Cookbook

Calculus (single- and multi-variable calculus).

  • The most important topic from calculus is the derivative and the way to find the maximum (or the minimum) of a function. This is because machine learning defines a functions and tries to go to its maximum (or minimum - depending on the problem) by using its derivatives. You should also understand integrals but you need only to have an idea of what an integral is at first and you can study more if you need it later (you need integrals for probability distributions). You do not care about the more theoretical concepts here (as if the function is continuous, etc.) so focus mostly on derivatives and optimization (e.g. finding the max or min) if there is available material on that.
  • Can't give you a reference here because I have only studied from textbooks written in my native language. Maybe Thomas' Calculus ?


Probability and statistics.


  • This is the most important requirement that you need to understand. Statistics is the basis of machine learning and this field is huge so don't get lost here. The concept of the probability and that of the uncertainty are the most important and you need to study to a point where you can understand probability distribution functions such as the Normal distribution , the Bernoulli distribution , etc.
  • Many excellent books here from where you can learn probability and statistics. Any of the following will help you to understand the topic really good:



  1. Introduction to Probability, Bertsekas and Tsiklis
  2. Probability Theory: The Logic Of Science, Jaynes
  3. Introduction to Probability and Statistics from a Bayesian Viewpoint, Part 1, Probability, Lindley
  4. Other Probability on-line free courses (such as Introduction to Probability - The Science of Uncertainty ) and textbooks might be good as well



  • Remember that if you find some topics hard enough you can skip them and come back later.

I am proposing a way to get you started. As you try to deal with more complex topics and ideas you might need to go back and study more that's the process of learning. At this point you do not need to focus on optimization methods as this will be overwhelming. If you want to study later then the Convex Optimization, Boyd and Vandenberghe is the way to go here. But you can start with basic machine learning text books and then find what you need to study next by yourself or by asking other people who know.

Second, the best sources to study machine learning (this is not an extensive list, many other excellent books are missing).

Machine learning.


  • Many great text books out there and I am only going to focus on those that provide a mathematical way to present thing and not on the cookbooks.


  1. The Stanford's Machine Learning  course is one of the best courses out there and the material taught in the class is freely available in the course's website. Check out the lecture notes (exceptionally good) and the Lecture videos from Andrew Ng.
  2. Information Theory, Inference, and Learning Algorithms, David MacKey , a free online textbook which provides many different ways that you can study it from page iv and on.
  3. Machine learning textbook, Kevin Murphy is the book that I am currently studying. It is excellent so far and most figure (maybe all?) have the accompanying MATLAB source code that you can download. I would recommend it to everyone.
  4. Christopher M. Bishop | PRML I have heard that it is a bit more of an advanced book and it will be my next stop after I finish the two previous books.


Machine learning applications


  • There are many tools out there and libraries/frameworks which you can use to do your job and you need to get familiar with those because they provide a good interface to the algorithms use.


  1. Python Data Analysis Library
  2. scikit-learn: machine learning in Python
  3. R for Machine Learning  
  4. And many more.....

*Different educational systems in different countries have different requirements so my answer tries to be as general as it can be.

Disclaimer: There are many different sources to learn the material you need. There are also many other GREAT books that I am not aware off yet. Please search other questions and answers and also visit a library to take a look at those books as they might not fit your needs. This is just a guideline on what you need to study and not from where to study it.