darb.ketyov.com

Caveat lector: This blog is where I try out new ideas. I will often be wrong, but that's the point.

Home | Personal | Entertainment | Professional | Publications | Blog

Search Archive

Loading...

26.6.12

General 2012 updates

Normally I don't do "life posts" like this, but a lot's been going on and I feel like sharing because dammit, I'm excited.

2012 has been a wild year so far: fatherhood, tons of cool new projects in my post-doc research, and a lot of outreach and science communication stuff.

So far I've had two papers published, with three more(!) first-author papers out for review right now. The two published are:


I'm really happy to see the brainSCANr project through to some form of completion with that second paper. I can't wait to share the other three papers coming out. Humbly, they're awesome. Well, two of them are; the third's okay.

I've also been doing some writing for the O'Reilly Radar (no, not the evil O'Reilly, the good one):


I also found out that I was a finalist for the AAAS Early Career Award for Public Engagement with Science. So close! But really cool to even be considered.

Also, a couple of weeks ago I was invited to attend Foo Camp which was a hell of a weekend. I really could (should?) do a whole post about my experiences, but wow there are a lot of really smart, talented people out there doing some great things. That's a really fluffy, non-specific sentence, forgive me, but I just don't have the mental energy to break all my conversations from that event down into details yet.

If you don't read anything else, please read this interview I did with the UCSF Clinical & Translational Science Institute about why I use social media. If you're reading this blog, I believe it will be relevant to your interests.

I've done a lot of science lecturing and outreach stuff this year as well. Deep breath. Here we go:

  • Spoke twice at swissnex San Francisco, totally independently; once about my actual research and once about my science outreach.
  • Gave a zombie brains talk (with a real human brain specimen!) at two local libraries for about 50-100 12-16 year olds.
  • Speaking at NASA Ames in a few weeks about my research and outreach.
  • Was the Keynote Speaker for the Annual Meeting of the San Francisco Neurological Society, also speaking about zombie brains, which was wild.
  • Speaking at the International Neuropsychological Society in Oslo in a few days.
  • Was invited by a University of Chicago neuroscience student group to give a talk there in October.
  • Will be speaking at the Allen Brain Institute in August about brainSCANr.
  • Presented a paper, “Medusa to Slake-moth: The Neurobiological Basis of Hypnagogia, Paralysis, Hallucinations, and Other Magical Abilities of Literary Monsters”, with my friend and mentor Roby Duncan at the International Conference on the Fantastic in the Arts (which we're hopefully turning into a peer-review publication).
  • I've done a lot of speaking at Bay Area companies about my data analyses with Uber.
  • Will be on another zombie panel at Comic-Con this year.
  • Will be speaking at the California Academy of Sciences to help kick off the Bay Area Science Festival in October again.
  • Did a radio interview with New Hampshire Public Radio about zombies last month.
  • Did an interview with SETI's radio show Big Picture Science about brain-computer interfaces in January.
  • Am one of Huffington Post's "30 high-profile neuroscientists who tweet".

There we go! That's 2012 so far. Holy crap typing all of that out is crazy. How the hell did I get here? Because I'm pretty sure last time I looked I was wasting time procrastinating and not doing my PhD research.

I wish I could convey to you all how amazingly surreal it is for me to be in this place in my life. Having gone from a... weird... upbringing, losing many people who were near and dear mentors to me, and getting kicked out of college because of my shite grades to... whatever this is, with family and friends and everything else, is astonishing.

I know this post may read like braggadocio, but believe me when I say this kind of thing is necessary for me. My close friend/mentor/dad-by-wager (long story) Roby once told me, when I was in the midst of a years-long bout of self-doubt and uncertainty, that every now and then I need to praise myself; rather than just focusing on what I'm not doing/accomplishing, I should also allow myself the guilty pleasure of basking in successes.

So here I am. Typing all of this out so I can say for myself "hell yes" to keep motivated and moving forward.

This will be it for a while, I promised. I will now return you to your regularly scheduled weird posts about brain damage and ridiculously unethical studies.

16.6.12

Defending Jonah Lehrer

(Edit 2012 July 30: Here we are six weeks later, and Mr. Lehrer has admitted to fabricating quotes and lying to a journalist about parts of his latest book. Obviously there is no defending this kind of behavior. This post was meant to illustrate some of the problems in neuroscientific thinking using a popular contemporary science writer. Sadly, it would appear I chose a poor horse to back on this one.)

This is a strange post for me to write because I admit I've ridden the anti-Jonah bandwagon before, advocating throwing Jonah overboard to quell the pop neuroscience storms.



Upon honest introspective reflection I admit that some of my anti-Lehrerism probably stems from righteous brain-nerd ego-driven indignation. Why does this dude get all the attention when he's not even a neuroscientist?! He's just a neuroscience roadie!

And that's not fair, and neither is all the shit he's getting.

He's taken a lot of flak lately, more so than I think he deserves. He's not a neuroscientist. He's loose with language and exaggerates or adds flourish. He's glib and fetishizes neuroscience.

But you know what Mr. Lehrer? I've come to realize something:

It's not you, it's us.

Really.

I would be more annoyed if Mr. Lehrer were intentionally distorting neuroscientific research to fit his needs, but in reading his works I don't believe he is. Full disclosure time: I've never read a single word of his books. They're not written for me. But I've been reading his columns on Wired and The Wall Street Journal and other outlets for years now.

So why is the neuroscientific community at fault for Mr. Lehrer's occasionally inaccurate scientific reporting?

Because our own house is in such disarray. Of course there are the well known issues in cognitive neuroscience, such as Vul's "voodoo correlations", "double-dipping" statistics in neuroimaging, and the dead salmon. Or our straight-up misunderstanding of basic statistics.

But some of our issues are more subtle. One of the main offenders living in our attic seems to be conflating the idea that because a brain region is active in one state--such as addiction--and in another task--such as mothers looking at pictures of their own babies--that babies are "addicting".

Or your iPhone is.

This makes about as much sense as saying that because I kiss with the same mouth-hole that I burp from, kissing and burping are essentially the same. (Note: they're usually not.)

Of course, if this logical inference were true, so too should be its converse. Maybe addiction is like having babies? Shouldn't it cut both ways?

And that's assuming that the "dopamine = reward" hypothesis is even true. Most people--neuroscientists included--take this as gospel truth. Of course dopamine equals reward! Dopamine neurons fire in response to rewarding stimuli, and the neurons "learn" to predict the rewards! Addicts' brains show activity in dopaminergic regions when shown images of drug paraphernalia. And on and on.

But--and this is apocryphal as no one would ever publish this--but from what I've heard the neurons that best predict reward values aren't in the dopaminergic brain regions. They're in the monkeys' neck muscles. Because the monkeys tense up in anticipation of reward. But few scientists would say that the neck muscle neurons are "encoding" or "predicting" reward, yet we make that fallacy all the time when we imbue neurons with that special computational power.

Oh, and by the way, dopaminergic neurons don't get any sensory inputs early enough to make a "decision" about the reward value of visual stimuli. In fact, they're probably encoding salience (relevance).



Which explains why drug users have increased activity when shown pictures of drug paraphernalia, and mothers pictures of their children, or even iPhone users pictures of iPhones versus Androids: because those things are more familiar and relevant to them.

To really hammer this point home, there is one disease we know of that is caused by the death of dopaminergic neurons: Parkinson's disease. It seems to me the clearest support for the argument that "dopamine = reward" would be seen in people missing most of their dopamine. Parkinson's patients shouldn't be able to experience any reward/pleasure because that whole system is obliterated.

Clinically, not feeling pleasure from experiences is known as "anhedonia", and a systematic review of the literature on Parkinson's and anhedonia in 2011 was inconclusive. In that review the authors found that, if anything, any signs of anhedonia in Parkinson's patients was likely caused by their associated depression.

This is very personal for me, as I watched someone very important in my life degenerate from Parkinson's disease. While an anecdote is not data, I can tell you he wasn't anhedonic. Instead he would just space out and stare all the time. Nothing seemed relevant to him.

It's not just popular writers who make these kinds of subtle scientific errors; we cognitive neuroscientists do it all the time as well.

It turns out some of our strongest neuroscientific results could very well be wrong. Or, at the very least, they're not nearly as cut and dry as they're often made out to be.

So how can we blame people like Mr. Lehrer for linking dopamine with reward when that idea has been one of the major results of systems and cognitive neuroscience of the last 30 years?

There is also the fallacy that poorly-defined (scientifically speaking) concepts such as "creativity" can be accurately studied neuroscientifically. How do you operationalize creativity, and more importantly how do you know that what you're seeing in the brain in response to your measure of creativity is the thing you think you're measuring? There's often no way to validate this.

When you ask something like "where is creativity in the brain" you assume that researchers can somehow isolate creativity from other emotions and behaviors in a lab and dissect it apart. This is very, very difficult, if not impossible. Neuroimaging (almost always) relies on the notion of cognitive subtraction, which is a way of comparing your behavior or emotion of interest (creativity) against some baseline state that is not creativity.

Imagine asking "where is video located in my computer?" That doesn't make any sense. Your monitor is required to see the video. Your graphics card is required to render the video. The software is required to generate the code for the video. But the "video" isn't located anywhere in the computer.

But if activity in that region increases as you're "more creative", clearly that's strong evidence for the relationship between that brain region and creativity, right?

Just like how when your arms swing faster when you run that means that your arms are "where running happens".

My point being, these errors are running amok in our own scientific house. Cognitive neuroscientists make these assumptions all the time.

Cognitive neuroscience grew out of experimental psychology, which has decades of amazing observations to link psychology and behavior. But with this legacy comes a lot of baggage. Experimental psychologists observed that we have the capacity for memory, attention, emotion, etc. and they sought to piece those phenomena apart.

With the advent of neuroimaging techniques, psychologists put people in brain scanners to see where in the brain behaviors "were".

But this is the wrong way of thinking about these concepts.

As cognitive neuroscientists, instead of asking, "where in the brain does this fuzzy concept occur?" we should be asking, "how can neurons give rise to behavioral phenomena that look like what we call creativity?"

Obviously I'm not saying that psychologists were doing things incorrectly. What I am saying is that we need to build upon what we've learned from decades of psychological research within a neuronal framework.

Not just stick people into an fMRI, press some buttons on a computer that say "analyze", and copy-and-paste the figures into a paper.

So Mr. Lehrer, keep up the interesting writing. Just... please be more skeptical of us. We don't know nearly as much as you give us credit for.

ResearchBlogging.org
Vul, Harris, Winkielman, Pashler (2009). Puzzlingly High Correlations in fMRI Studies of Emotion, Personality, and Social Cognition Perspectives on Psychological Science DOI: 10.1111/j.1745-6924.2009.01125.x
Nieuwenhuis, Forstmann, Wagenmakers (2011). Erroneous analyses of interactions in neuroscience: a problem of significance Nature Neuroscience DOI: 10.1038/nn.2886
Kriegeskorte N, Simmons WK, Bellgowan PS, & Baker CI (2009). Circular analysis in systems neuroscience: the dangers of double dipping. Nature neuroscience, 12 (5), 535-40 PMID: 19396166
Redgrave P, & Gurney K (2006). The short-latency dopamine signal: a role in discovering novel actions? Nature reviews. Neuroscience, 7 (12), 967-75 PMID: 17115078
Assogna F, Cravello L, Caltagirone C, & Spalletta G (2011). Anhedonia in Parkinson's disease: a systematic review of the literature. Movement disorders : official journal of the Movement Disorder Society, 26 (10), 1825-34 PMID: 21661052

8.6.12

In defense of frivolities and open-ended experiments

I've got a new post over on the O'Reilly Radar that's a riff on my previous post defending basic scientific research.

In it, I ramble about fatherhood, the Pyramids, Nightcrawler, Python, Louis C.K., and... wait. What?

Now that I've written that out I'm not sure you should go read it at all...

4.6.12

Data munging

Yet another post inspired by an answer I wrote on Quora. This one to the question, "Data Science: What are some must-know tricks in field of data science that most people are oblivious to?"

This is a weird aspect of my research that I don't think most people even know is an issue, which makes it so interesting to me.

Data munging.

In my opinion the "trick" of data munging is the most surprising skill any "data scientist" must learn. It seems so innocuous: get data from one dataset to match up with, or in the same format as, data from other datasets.

For anyone who doesn't deal with a lot of data it's simple: data are data. That is, data are either spreadsheets with some numbers in ordered columns and rows or "bits on a computer or something".

Anyone who's worked with lots of data knows how nightmarish getting data from disparate datatsets into a workable fashion can be.

Yes, in the end it's all just bits, but fitting together all the pieces can be maddening.

YOU'RE ALL THE SAME, DAMMIT! PLAY NICE.
Some data are stored in nice *SQL or newer databases with wonderful metadata. Those are your JSON, XML, or even CSV formatted data. People will complain about XML vs. JSON, and redundancy of information or what have you, but given some of the alternatives I'll choose redundancy, thank you.

Sometimes you've got some Excel files to work with... okay, that's not terrible, but again, given some alternatives, sure, let's go with Excel.

Sometimes you've got proprietary formats that you'll need specific software for in order to get at the data. Again, not ideal, but someone out there has probably made a conversion tool, or you can just suck it up, get the proprietary software, and output the data to a nicer format.

Sometimes, some organizations (::cough:: US government ::cough::) have made their data public (yay!) but stick it into not-easily-machine-readable formats or layouts, such as PDFs. So sure, you can open the PDF and see how the data are laid out by column and row, but try automatically pulling that into your analysis software or copy-pasting it into a database and see what happens. Grrr...

But the bad days... the bad days are when you need to crack into the data because it's locked away in some obscure format from some obscure software that hasn't been used in decades. That's where you need to know about endianness and read that shit in bit-by-bit and go digging around for format specs and whatnot.

And then there are issues of data transmutability: are there non-printing Unicode characters in your data file? HA! Surprise! Your data munging algorithm is now borked until you code a bunch of special clauses.

What about date and time formats? Is your serial timestamp stored in Unix time? Because that begins in January 1, 1970. What about Excel dates? Those begin in January 1, 1900, unless it's on a Mac, then they begin on January 1, 1904. Got that? Was there a Leap day? Leap second?

(see also What are all the known dates/places it's impossible to have been born?)

What time zone was your data from? If it's the US, don't forget daylight savings time! Unless your data are from a state that doesn't observe daylight savings, or they did, but don't now. Or won't. Or... just be consistent illogical humans!

And sometimes (not often) it's important to know if your data comes from a system that uses zero-based indexing, or one-based indexing (boooo MATLAB). For example, in SQL, you can just join the tables below to quickly find out that Rafferty works in Sales.
But what if your Department Table didn't have an explicit DepartmentID, but instead the ID was coded by position in the array such that Sales is the first element, Engineering the second, and so on? You need to know if your explicit DepartmentID in the Employee Table is based on a zero index system such that Sales would be coded 0, or a one-based system where it would be coded 1.

The worst is when you have to try and infer the answer based upon the data distributions themselves. If you think it's a zero-based index, but find out that, based on that assumption, that no one works in sales, which you know can't be true, then you have to infer that it's a one-indexed array. At least, you hope that's right...

::phew::

Glad I've got that out of my system.

So anyway, "data science" gets a lot of sexy attention and people think you spend your time like this:


when in reality it feels like a lot of time is spent like this: