Disqus - Latest Comments for dtunkelang

Re: Asking the Right Questions: Query Expansion in Enterprise Search

Daniel Tunkelang — Tue, 13 Aug 2019 17:57:33 -0000

Martin, as always I appreciate your posts. And I have great respect for Liz and Tony. As for our differences on the uses of query expansion, I think we may have different applications in mind. I'm specifically focused on using it to increase recall, rather than to increase precision. A lot of IR folks focus on the latter, e.g., query expansion using pseudo-relevance feedback. So I'm not sure we disagree -- we may have just been talking past each other.

Re: 5 Things to Know About A/B Testing

Daniel Tunkelang — Tue, 11 Sep 2018 02:19:05 -0000

Nice post! For those specifically interested in applying A/B testing to search, I encourage you to read my post on the subject: https://medium.com/@dtunkel...

Re: 8 Useful Advices for Aspiring Data Scientists

Daniel Tunkelang — Fri, 25 May 2018 16:38:43 -0000

My above advice feels pretty dated. Today, an aspiring data scientist should be learning Python, Scala, and Spark.

Re: How To Do Startup Technical Due Diligence – code.dblock.org | tech blog

Daniel Tunkelang — Mon, 30 Oct 2017 11:20:21 -0000

Great post! One question I don't believe you addressed: what do you feel is the required specificity of expertise on the part of the person doing due diligence? For example, do you need an AI expert to evaluate an AI startup? Given the scarcity of people who are qualified and available to perform technical due diligence in general, adding specific expertise as a requirement can harshly limit the options or make the process significantly more expensive. But sometimes it seems necessary.

Re: What We Learned Analyzing Hundreds of Data Science Interviews

Daniel Tunkelang — Tue, 16 Aug 2016 17:50:11 -0000

Thanks Roger!

Re: What We Learned Analyzing Hundreds of Data Science Interviews

Daniel Tunkelang — Sat, 13 Aug 2016 14:45:16 -0000

Interesting post. But your information about LinkedIn is a bit dated. I only ran part of the data science team -- the folks with more of an engineering bent who built data products -- and I stopped doing that over 3 years ago. The company has had a few re-orgs since then, cf. http://venturebeat.com/2014...

Re: Is It Time For Google To Rank Paid News Content Better?

Daniel Tunkelang — Sun, 17 May 2015 13:48:05 -0000

Interesting idea. But I think you're eliding over a big differences between news and music / videos. News has a high degree of fungibility -- if I can't read an article that's behind a paywall, it's likely that I can legally read something almost as good (yes, I realize that's a subjective measure) that isn't behind a paywall. For music and video, there's no comparable legal substitute unless the rights owners opt for one. Your proposed solution may work, but I don't think it's obvious that what works for music / video will generalize to news.

Re: The Case for More Women in Data Science

Daniel Tunkelang — Mon, 20 Oct 2014 09:56:34 -0000

Claudia, thanks for having the courage to publish a piece like this -- I know it's not your usual publication material.

To those of you not familiar with Claudia's work, I encourage you to look at her papers in top-tier computer science journals and conferences:
http://scholar.google.com/c...

Or, if that's too technical for you, at her more accessible material on SlideShare:
http://www.slideshare.net/S...

Re: The Myths & Realities Of The EU’s New “Right To Be Forgotten” In Google Works

Daniel Tunkelang — Sat, 17 May 2014 10:47:37 -0000

Outstanding and thorough explanation, Danny. That said, it only reinforces my outrage at the European Court of Justice for taking such an irrational approach to suppress the freedom of expression under the guise of protecting privacy. I'm no Google fanboy, but in this case the company has been acting as a voice of reason and sanity against governments that disgracefully haven't learned that history cannot and should not be forgotten. Not to mention the inconsistency of going after the search engines but not the original publishers. Ah well, let's see them try to enforce this madness.

Re: Google is researching ways to make encryption easier to use in Gmail | VentureBeat | Business | by Harrison Weber

Daniel Tunkelang — Wed, 23 Apr 2014 14:42:30 -0000

Perhaps I wasn't clear. I'm talking about processing encrypted email in the browser / app client in a way that retrieves targeted ads but doesn't share the unencrypted email back to Google at all. At most it would share a few keywords or some kind of vector that minimizes disclosure of the email content. This is technically feasible -- but I have no idea if Google or anyone else is doing this. Which is why I'm curious to see a reference for Bob Bigellow's statement that Gmail ads already work client-side.

Re: Google is researching ways to make encryption easier to use in Gmail | VentureBeat | Business | by Harrison Weber

Daniel Tunkelang — Mon, 21 Apr 2014 15:45:39 -0000

Interesting, I didn't know that. Could you point me to a reference?

Re: Google is researching ways to make encryption easier to use in Gmail | VentureBeat | Business | by Harrison Weber

Daniel Tunkelang — Mon, 21 Apr 2014 13:06:15 -0000

Actually, Google -- or any other ad-supported mail provider -- could implement ad targeting in a way that processes the page client-side, without ever storing the clear text of the emails on its servers. Probably a lot more complex, but that may ultimately be the price of secure messaging. Unless consumers decide that security is worth paying for.

Re: Forgotify: The Tool for Discovering Spotify's 4 Million Unheard Tracks

Daniel Tunkelang — Sun, 02 Feb 2014 22:55:10 -0000

Interesting -- trying it out now. Any idea how much of the song do you have to listen to for it to qualify as "played"?

Re: The Bay Area’s 1 Percenters

Daniel Tunkelang — Sun, 27 Oct 2013 17:21:52 -0000

Let's try to move beyond personal experience and bring some data to the conversations.

Palo Alto public high schools have class sizes above the state average:

http://paloalto.patch.com/g...

In Mountain View, overcrowding has become enough of a concern at a local public elementary school to motivate a policy study on alleviating it:

http://publicpolicy.stanfor...

Is it possible that some of the "stampede to enroll students in upscale private academies" is simply overflow from overcrowded public schools in an area that has been struggling to manage an increase in the student population?

Re: The Bay Area’s 1 Percenters

Daniel Tunkelang — Sun, 27 Oct 2013 16:37:29 -0000

I think it's pretty accurate to say that Hanson is pretending that people like me don't exist by painting a one-sided picture. That's his prerogative, much as it's mine to call him on it.

Re: The Bay Area’s 1 Percenters

Daniel Tunkelang — Fri, 25 Oct 2013 13:23:44 -0000

And here is someone who has responded to Vic's claims with some data. It's pretty opinionated, but so is Vic. In any case, everyone is entitled to their own opinions, but not their own facts.

http://www.cjr.org/the_audi...

Re: The Bay Area’s 1 Percenters

Daniel Tunkelang — Thu, 24 Oct 2013 09:17:05 -0000

Vic doesn't offer data or even define a "stampede", other than implying that he talks with friends who live in Silicon Valley. So it's hard to evaluate the truth of his unfalsifiable and unsubstantiated claim.

What I do know is that that the hot real estate market on those areas is driven in significant part by "1 percenters" wanting their kids to be able to go to those local public schools -- a big part of property value comes from the associated school district. That argues more for a stampede into the local public schools than out of them.

Re: The Bay Area’s 1 Percenters

Daniel Tunkelang — Thu, 24 Oct 2013 00:42:15 -0000

Victor, I have to wonder if your preconceptions are leading you to only see what you want to see. My wife and I, like many Silicon Valley parents, send our child to public school even though we can afford to send her to private school. We deliberately chose a school that was majority Hispanic. Partly because we want her to learn Spanish (it's a dual-immersion program), but also because we want her to get to know people with different backgrounds. Pretending that people like us don't exist makes me wonder if you're even trying to make an objective analysis. Don't caricature people just because you disagree with them politically -- it's petty.

Re: What Mugshots Mean For Public Data

Daniel Tunkelang — Sun, 06 Oct 2013 15:23:29 -0000

This debate only reinforces what Jim Adler has been preaching for a while: we need to regulate data use rather than data access.

http://jimadler.me/post/266...

Re: a16z

Daniel Tunkelang — Mon, 19 Nov 2012 22:31:16 -0000

Congrats Chris! Looking forward to seeing you here in the Bay Area. And congrats to a16z for landing you!

Re: Hubris and the Data Scientist

Daniel Tunkelang — Mon, 05 Mar 2012 17:52:57 -0000

Paul, I wish the discussion had been recorded. As I wrote in my blog post, the question proposed was absurdly Manichean: if you had to hire your first data scientist and could only hire one, would you pick a domain expert or a machine learning expert? Most of the room disagreed with the premise of the question, but the debaters made the most of it by taking extreme positions and defending them with gusto. It was a lot of fun, with enthusiastic audience participation and the debaters exploiting their inside knowledge of their opponents’ work histories.

I'm not denying that we data scientists have to watch out for hubris. But the community is not as extreme as to not find a place for domain expertise. In fact, much of our work should help to both objectively validate and maximally benefit from domain expertise.

Re: Strata 2012: Is Privacy a Big Data Prison?

Daniel Tunkelang — Sat, 03 Mar 2012 16:51:29 -0000

Loved your session, even if it was way too short. Eager to pick up on the question of whether it makes sense to consider inference illegal or unethical and, if so, how we don't devolve into creating a class of thoughtcrime -- which, ironically, would be the ultimate government invasion of privacy.

Re: Don't write on the whiteboard by Joseph Perla

Daniel Tunkelang — Tue, 10 Jan 2012 16:45:22 -0000

Joseph, thanks for all the traffic this morning! :-)

I hope this is what you mean by a "Dan Tunkelang type problem":

1) It is a real problem that has come up in the couse of developing production software.

2) It does not require any specialized knowledge — just strings, sets, maps, recursion, and other basics that are covered in a first- or second-year undergraduate course in computer science.

3) The code is non-trivial but compact enough to use under the tight conditions of a 45-minute interview, whether in person or over the phone using a tool like Collabedit.

4) The problem is challenging, but it isn’t a gotcha problem. Rather, it requires a methodical analysis of the problem and the application of basic computer science tools.

5) The candidate’s performance on the problem isn’t binary. The worst candidates don’t even manage to implement the fizzbuzz solution in 45 minutes. The best implement a great solution in 10 minutes, allowing you to make the problem even more interesting. Most candidates perform somewhere in the middle.

Re: Duck Duck Go

Daniel Tunkelang — Thu, 13 Oct 2011 11:15:32 -0000

Brad, congrats to you and to Gabe! Was an early user of DDG -http://thenoisychannel.com/... - and happy to see folks trying to make web search better. It's an uphill battle, but I'm sure you all know that first-hand!

Re: Whither TechCrunch?

Daniel Tunkelang — Thu, 08 Sep 2011 11:03:23 -0000

As my friend Rob Gonzalez says, structured data repositories are a public good that no one is ever willing to pay for. Freebase has always sounded great in theory, but has never lived up to its promise because it can't deliver on coverage.

http://thenoisychannel.com/...