kahnjw

1. A sufficiently large distribution will contain outliers. 2. Where is Madoff now?

>It actually gave me the impression that avoiding the ire of the SEC is relatively easy if you set your mind to committing fraud without getting caught.

I assure you, it is only possible to avoid risk if you're a small fish and extremely careful. For instance, you have a good chance of getting away with skimming a neighborhood stock fund you setup. If you are managing or receiving millions of dollars in investment, you can bet anything fishy will show up on the SECs radar at some point. Most people with large sums of money don't get where they are by being stupid with it, and if they are the victims of fraud, they know where to go. While the SEC can't bring criminal charges, they have top notch investigators. Once they smell the fish, they won't stop until the entire scheme is unraveled.

I can't speak directly to the observation you made, but I personally know a high up SEC investigator, and he obsesses about the most minute details of the cases he works on and lives to take on the toughest cases possible. My guess is distribution of cases the SEC prosecutes are very heavy tailed. There are many dumb cases that are very easy to find and prosecute that involve 10s to 100s of thousands of dollars. Think skimming, ponzi schemes etc. The more sophisticated and larger cases are also more risky (for the fraudster), harder to execute, and much more rare. That might explain your observations.

All of those Siri can do maybe half the time. In my experience, it's a completely unreliable service. Sort of embarrassing when your voice assistant can't perform the most basic possible task on an iPhone: locking the screen.

>>...watch as the last bastion of python gets overrun

That is a naive point of view. The python community has been growing for decades [1]. Even if it is not growing as fast as javascript community, the growth is sustainable, and probably won't slow until the companies that rely on it for core parts of their infrastructure are gone (that could be 5 decades). Google, youtube, instagram, and countless other enormous corporations with trillions of dollars in combined market cap are dependent on it. Python isn't going anywhere.

[1] https://trends.google.com/trends/explore?date=today%205-y&q=...

Aren't you sort of glossing over the fact that he is in high up machine learning position at a company that specializes in recommender systems? Doesn't that by itself increase the likelihood that he deeply understands implicit and explicit matrix factorization?

I am a good ways through my masters (second CS degree, first specializing in ML), and the more I learn, the more I realize that on any given topic, there is no guarantee the PhD in the room has the most expertise. Machine learning is a broad field that contains many subfields, methodologies, and many applications. It is a bit like computer systems or software engineering: nobody knows it all, people who are experts have intimate knowledge of a specific subset of the field. Of course, you can more around over time, but it takes years to build up expertise in even two or three subfields of machine learning.

Side note: sounds like we do similar work. I work at Vevo, also do a lot of matrix factorization to learn latent factors of items such as artists, videos, etc.

Agreed. Scientific computing has centered on two languages for well over a decade. I don’t think JavaScript is going to dethrone python as the interpreted language of choice because it can now push computation to the GPU via a well designed API.

And the full force of the entire machine learning research community (except for a few still on Lua/Torch).

Agree with 1. Disagree entirely with 2. I have two MacBook Pros (one from work) and the keyboards both have issues. One is newer and they tried to fix it, didn't work. The touchbar is a gimmick. It's decent on small workloads, and of course is fairly useless on heavy workloads. But none of that surprises me. They push the limits of what is possible with hardware design. The worst part is price. These macbook pros are close to $4k each. Go ahead and look what you can buy for $4000 if you aren't buying apple. You can get laptops that are much cheaper, much faster, somewhat extendible, and _well designed_. This is probably the last macbook i'll buy unless something changes.

Apple definitely is dropping the ball on its professional customers. These machines are great for browsing the web or writing some code (assuming you have an external keyboard), but anytime you need to do some actual computation, you need an EC2 instance or a different workstation. The DL/ML community is more or less left out unless they use an external GPU chassis (which is trash).

I get that they are not trying to make workstations or servers, in any case, higher end chassis should be extendible, at least if you want professionals to use your products.

Anyone dropping ~10k on an iMac Pro either has more money than they know what to do with, or they're a fool.

Agreed 100%. I did a bunch of js work about 3 years ago, used tons of promises. Then started a new job using Scala. The futures api in Scala is exactly what the author advocates, and it is definitely better for the reasons he gives.

Have you heard of the precision/recall tradeoff?

In all seriousness, the problem with insisting on comparing the US economy and the economies of China and India is that they are at very different points in economic development. The US developed huge swaths of infrastructure post WWII, and that infrastructure is aging out and it clearly shows. India and China on the other have cheap and widely available labor to develop new infrastructure.

What happens when that labor wants better pay and safer working conditions? Things will naturally slow down, and it will be hard to maintain all the infrastructure that was built during growth years. The same applies to healthcare, yes it is egregiously expensive, and life expectancy is falling slightly. On the other hand, comparing the US health care system to India's is silly. Health insurance might have high costs, but the level of care you get is far better than what you get today in India. Ten years from now, that might be a different story.

It's pretty efficient if you're in the 1%.

>> The first order predicate calculus (first order logic) is Turing-complete

Brainfuck is turing complete, does that make it as powerful as a general purpose programming language such as Java?

I'm not criticizing first order logic, it has use cases no doubt. On the other hand there are many cases where codifying knowledge into a we formed KB isn't practical. Some of these use cases are better suited to probabilistic approaches, deep learning, and some are unsolved. That pretty clearly puts limits on the power of FOL.

Also, thanks for mansplaining turing completeness.

Could not have said it better. Hinton is one of the most important contributors to the field, but giving him all the credit is a disservice to others who made important contributions. Probabilistic modeling was popular way before DNNs got big. At that point it was already clear that logic based inference was only so powerful, and different methods were necessary.

Seems like these articles are becoming more common. Someone claims new meta-ai alg will change everything. In reality, just another team's hyper-param optimization setup.

The attack relies on side channel attacks, specifically Flush+Reload, which uses timing to find accessed memory locations in another program, as stated in both the paper and this summary. It is definitely relevant.

Non amp version, sorry about that

1. Third paragraph: "twice a week" 2. They registered multiple accounts to get around the limit. That removes almost all doubt that the submitter knew he/she was cheating

I didn't even realize this happened. Thank you for posting. Sucks because now I have less respect for Andrew.

I don't understand why you'd want to cheat for a competition like this? I get it, people cheat all the time, but the field of machine learning is built on a foundation of open and shared research, and trust.

You might be able to find a credit Beaureu that does not report to equifax. Might be better off calling your representatives in Washington and telling them we need better consumer data protection regulations. Good luck.

Experian != Equifax

If we're talking about economic incentive, having no downside for not doing thing A is equivalent to having no incentive to do thing A. Re: opportunity cost.

Have you heard the term moral hazard?

Exactly. And congress can fix it.

GLBA was enacted in the 90s and is a vague set of guidelines for financial and related industries to follow. I'm talking about real system regulation, down to the level of protocols.

Thats exactly what I'm saying. There is near zero incentive for congress to enact consumer protection legislation.

You're oversimplifying. This has more to do with politics than SE technical competence. Banking software and systems are regulated end to end, because banking systems are literally the backbone of the global economy.

Equifax? Not so much. This exposure hurts consumers, but it doesn't even put a dent in the economy as a whole. That is why these firms are allowed to operate with such shitty security. If they get hacked, whatever, just a few hundred million customers data exposed to identity fraud. It takes a chunk out of Equifax stock but they'll probably survive. What is the incentive for congress to enact laws regulating corporate handling of consumer data?

Put it this way. Who do you think is louder in Washington: consumer advocacy groups or corporate lobbyists?

If you really are a talented machine learning specialist, you don't have too much to compete with in San Francisco. A lot of "ML engineers" and data science people in SF dont actually have the background or expertise to call themselves career machine learning specialists. They take free courses and started using tensorflow once it blew up. I know because I was one of those people (fwiw im going back to school to actually learn the fundamentals now).

Also, I know people in the situation you describe, and its a grind like every other bonus incentivized gig on wall street. People may tell you otherwise, but they're probably sugarcoating. I know a few people who burned out early and started their own companies, and a couple who are continuing the slog for huge bonus payouts. The no free lunch theorem holds true.

For me, I'll finish up this degree and see where things are headed. I'm definitely not set on SV or WS yet, but it'll probably be one or the other.

Shameless plug: I recently wrote a similar piece exploring pattern matching in Javascript http://jarrodkahn.com/posts/6/slug/