New online master's degree to train the data scientists of tomorrow
ischool.berkeley.edu
ischool.berkeley.edu
They may need better data science behind their marketing.
I'm sure some firms will really like these degrees, but if you're a company (like Google, for example) insisting on a specialized degree, why not aim for PhD level? The income differential is marginal. In Chicago, PhD quants for funds generally start out around $140-150k. I'd argue those ppl are going to be more productive researchers/data scientists than what can be produced by an online program, due largely to value of a research oriented degree versus a skills oriented degree like this one. Even if there is a project component, it isn't the same as writing a thesis. I did an M.Eng way back, trust me: it just isn't the same as banging your head against a research topic for a couple years.
Also, for what it is worth, I found the target audience for one-year financial engineering degrees (Goldman Sach, etc) didn't respect the degrees as much as PhDs & MS w/thesis. They generally regarded it as a 5th year of university.
Criticism aside, I'm sure this is a great skill builder and ML is fun thing to learn. Not for $60k, though.
I like how they're still marketing that McK data science report. "Hadoop everything"
I do disagree with this snark, though:
> This was back when "financial engineering" was the buzzword... that worked out well, didn't it?
My M.Eng (Cornell ORIE) was not in FE, though my department offered it. Their (admittedly biased) response to this line of thinking: if we had more financial engineers, we would have had people who actually understand the instruments that were being traded.
No doubt the people who created exotic Mortgage-Backed Securities were FE types... if only Moody's and S&P employed some as well, perhaps they wouldn't have been rated AAA. Then again, I'm assuming incompetence rather than malice; the ratings agencies did have incentives to lie.
EDIT: I did mine at CU back in 2003 in applied, not FE. I had many friends in FE, most all went into credit. I worked in trading for six years before quitting for a PhD. I do not place any value on an FE degree; looking back at the curriculum they offered, it is obvious that they were thinking the wrong way (the credit models they were teaching were complete shit and they had no concept of micro-structure; ironic given Maureen O'Hara teaches at the Johnson School).
EDIT2: The non-FE profs were and still are very awesome and remain good friends. Did you have Henderson?
Henderson is one of my favorites; I just went back for my 5-year and he was just so wonderful to talk to.
The FE curriculum never really interested me; I took OR methods in FE as an undergrad and the professor (a surfer-dude postdoc from UCLA, Will Anderson) convinced me that the efficient market hypothesis was mostly right.
After that, FE seemed a little... well, in the words of my classmate Ryan, "like looking at the surface of the waves to see if there are whales humping."
IMO, continuous time finance is a flawed model. Academics always rattle off theorems based on assumptions that do not hold across all time scales. I've worked with traders lacking any formal education that have a better understanding of the market than someone like Protter will ever have. Trading isn't about investing; it is about capital flows.
That said, I don't miss the job (only the $).
narenl 5 hours ago | link [dead]
IMO This is not targeted towards individuals applying on their own.
I went to a walkabout of a similar online education company's office and asked them about the high cost for an online degree. The answer I received was
1. there is demand for this and
2. Most of the demand comes from people in the military serving in remote locations or people working inside other large organizations which foot the cost. and
3. They provide online infrastructure to courses of schools like UC-B, UNC etc and the colleges set the price so as not to dilute their "brand" because the online degree does not mention the fact that the degree was obtained online (this could have changed).
All in all, my initial shock was a bit tempered after hearing the realities involving all 3 parties : the school, the student and the online enabler.
The $60,000 price tag must come with a guarantee that you'll be making the proposed $110,000 - $130,000 salary range for at least 5 years. Otherwise, wow.
The best way to learn these things is to just dive right in. If one need's human interaction, the community is easily within reach (at a much lower cost).
Maybe one could argue that the "networking" is worth the price tag. Still, I get these "data meetup" emails about 5 times a day, which I could easily go to for networking.
I'm just not sure this makes sense. While tuition has gone way up at Berkeley, most science and engineering degrees are considered academic, rather than professional degrees, so the fees are considerably lower.
http://registrar.berkeley.edu/Default.aspx?PageID=feesched.h...
So yeah, law or business school tuition+fees (professional program) are between $50-$60 a year, but academic programs (which includes engineering) is less than half that. A lot of this comes down to whether data science will really be a "professional" degree with high earnings.
Truth is, it might, I'm not ruling it out. I got an MS in Industrial Engineering from Berkeley after a math degree, hoping I could get something practical. I had some good experiences, but I also spent what was to me a depressing amount doing proofs about convex sets and stochastic processes. I probably shouldn't complain, that's what an academic degree program is. Maybe something like a professional program would have been much better for me.
Another question is whether holding this degree is valuable independent of what you learn. I know that may sound silly, but it makes a difference. Suppose you were allowed to study law courses on coursera. How much would it be worth it to get to say your degree was officially from Harvard and now qualify for the bar, even if all you did was quietly watch the videos and do the homework? More than $60k, I'd say.
Could the same be said for data science? You've watched the identical videos and done the homework... how much of a premium would it be worth to say you got an MS degree online from Berkeley? It would be worth something sure, but not as much as the law scenario. There's no "data science bar" that can prevent you from practicing, and there are so many different acceptable degree paths to becoming a data scientist. And while some may reasonably dispute this, I have found high tech to be more concerned with what you know than where you learned it.
All in all... sounds like a great degree, but 60K definitely gives me pause.
Its funny to me that state schools, including Berkeley, used to have reasonable tuition up until the early 1980's, until the baby boomer voters in each respective state demanded lower taxes and the politicians obliged by cutting state tuition assistance year after year. When the state paid 90% of a student's tuition, the colleges didn't have the option of constantly jacking their price up. Suddenly, when it was student debt paying the bills, the colleges stopped doing anything to cut the bloat.
Thank you Berkeley for again reminding me what a scam the academic institutions have become.
I think you make a crucial point. Even if you were eligible in both situations to sit for the bar (say by virtue of California's looser education requirements), the Harvard name would be worth a heck of a lot more than $60k, at least assuming you were curved against other Harvard students in your examinations. The value of a law degree, at least in many circles of the legal profession, is 50% name brand, 40% sorting within the class, and maybe 10% education.
A few fields take this trait of education to a counterproductive extent (banking, consulting, law), but I think the unfortunate truth is that for most fields, the value of a degree is more than 50% in branding and sorting, rather than education. Even in engineering, outside some progressive places, I wouldn't put it much below 50%.
Part of the thing with Harvard, is getting in is so hard that if i'm interviewing a candidate half the job is already done. But if anyone is allowed into the program (and why not?) then that filter is gone... so the name while still having value as a great program loses the value as a filter.
Perhaps rather than offering snappy responses with negative tones, you could offer something constructive to the discussion?
Say, what you think the skills required for day-to-day work as a data scientist are, and how you'd suggest someone develop them.
Perhaps also what you think the best approach is to credentialing your learning--grooming a pedigree--if neither Kaggle nor a degree program are good approaches.
I'm not a data scientist, but I work with them very closely as an engineer and I've considered going down the same path. When I talk about data scientists, it's not a reference to any of the following:
> Engineers working with big data technology, like Hadoop, Storm, Kafka, who are essential but often uninvolved in model construction and evaluation.
> Analysts who develop models, then hand them off to engineers/IT to code them up (or keep them in Excel spreadsheets).
Instead, I'm thinking about someone with a specific background. They likely have a PhD, since that's an excellent way to experience the "ask-explore-code-test-present" workflow needed to answer an interesting question with real-world implications. The strong academic background is not necessary, but it greatly reduces friction during the research workflow (since you've spent 3-4 years in it). I'm getting a MS and working hard to make it as research-oriented as possible, fwiw.
This person also has a strong foundation in applied math. They might have worked on signal processing questions, applied algorithms for learning Bayesian network structure to proteins, or thought about the transition from Hopfield networks to RBNs or whatever awesome deep learning stuff is going on nowadays. A guy I respect described this quality as that of "a traveler," someone who can understand advanced work in a number of disciplines in addition to their specialty.
This person is an engineer. They learn languages easily, understand algorithmic complexity and think about the complexity of their models. They don't have to be Linus.
Finally, the person is forward-thinking. They understand that questions are motivated by business needs, and that answering these questions can have serious implications for the company or its partners. I should channel patio11 here!
Anyway I'm obviously very opinionated about this, but it's just one opinion. I'm happy to discuss this more with anyone who's interested, though--contact is in my profile.
https://news.ycombinator.com/item?id=6060821
There are two pieces Kaggle can't help you with: working through the full research cycle and developing performant models. It also emphasizes the wrong goals (for example error minimization is almost never your primary goal), but I need to work at some point and have spent enough time in this thread, so I'll skip that. :P Email me if you want to discuss, though.
Anyway Kaggle can't help with the full research cycle, since you're not identifying a relevant question yourself (this is surprisingly hard) or presenting your answer to others. The latter is hard for any route, since you really only encounter that type of volume in industry.
$60,000 is laughable. The justification based on salaries in California is laughable.
Unless a relocation package to California from anywhere there is internet is included in the fee?
Even then......
I grew up in SoCal and miss it terribly, but my cost of living in Chicago is a tiny fraction of what it would be in SF and my income isn't behind California norms.
Maybe if I was 24 and looking for experience... I suppose you can apply the same argument to finance in Manhattan.
I think their pricing may come from the fact that data science is the 'hot' field right now, so I suspect they'll be capitalizing on the corporations that will start pumping money into training their employees in data science. So I wouldn't be surprised if you start seeing some forture 500s covering the cost of this for a new hire.
First, the field is not professional like law, medicine, or even some parts of engineering. So, there's no licensing, board certification, recognized professional continuing education credits, professional job performance peer-review, legal liability, etc. Instead, you can just say that you have a Master's in data science and know some programming, database, statistics, etc.
Second, the degree isn't really a direct approach to business or entrepreneurship. So, the degree is aimed at making a person an employee. This means that somewhere there must be an employer including one ready to create a job, recruit someone for that job, and pay $120,000+ a year for the person.
Now, just who is going to create this job, e.g., put it in their budget and partly bet their career on it? And just why? I mean for what the program taught in programming, database, statistics, something else? And where will the real money actually come from, i.e., who with real P&L responsibility will actually cough up the $120,000 a year plus benefits, office space, travel, etc.? Or, let's think about the $120,000 a year: Ballpark, the full cost stands to be twice that, $240,000 a year. After two years on the job, maybe the person has actually delivered some value or is ready to start. So, the two years is $480,000. Heck, guys, even in Silicon Valley, that's a large seed round or a small Series A for a whole company and not just one employee slot!
I don't know but can ask: Are there some people at Berkeley smoking funny stuff?
The problem with data science is that it is incredibly hard to teach anyone Linear Algebra, Probability, statistics in 9 weeks. Sure, I can hand wave all that and then teach you a bunch of machine learning algorithms. All you get at the end of it is people who claim they understand it intuitively and don't need the math. Except that mathematical intuition builds up accumulatively.
It is easy to see this in interviews; you can see folks who are really good at drawing pretty pictures to explain say PCA. They have no clue when not to use such a thing. It makes no intuitive sense to them why PCA breaks down when there are outliers. If they can't draw a picture of it, it is difficult for them to comprehend.
Harvard's online masters degrees are closer to $20K all in.