The verbs of Data Science

The kerfuffle about what is Data Science and who are Data Scientists comes down to nouns and verbs.

Data Science and Data Scientists are nouns. Most nouns are abstractions, buckets that make communication easier. Nouns are short-hand, but they can slippery (some people do not tolerate ambiguity well.). Nouns become more useful (and more powerful) when paired with verbs.

Here are sample data science verb-noun pairs:

• Fit models.
• Create data products.
• Communicate results.

Each one of those general verbs can be made crisper with more specific verbs:

• Pipe data into Vowpal Wabbit .
• Code the backend to a recommendation widget for a website.
• Post slide deck.

I much rather have a discussion about verbs than nouns.

Skills are more important than interests

It is more common to see a "research interests" section than a "research skills" section on a curriculum vitae (CV). This is a reflection that it is easier to read review papers than learn Unix. A paper won't tell you that you don't understand it. On the other hand, Unix won't "go" if you are not precisely correct.

I rather work with people who view their professional work as a craft. As a craftsperson, they are proud of their toolbox. More pragmatically, I want an assurance their toolset is current and relevant.

Visualization baked into analytic processing

Current statistical analysis software assumes a matrix mindset. Easy for computers and the associated math but hard for most people, including me.

I am a visual explorer. When faced with a new problem, I like to draw to think. I envision a successful outcome before I start.

Current software doesn't help that process.

I imagine a different kind of statical program that pictorially represents raw data and exploratory analysis. Raw data could be icon-based, little homunculi with group tags. Groups could be represented as circles, mean differences represented as the distance between circles and variances represented as diameters. Confidence intervals could be added as concentric circles.

Visual quantitative reasoning is an important, but under taught skill. It is often the last step, often use only used to interpret finished graphs in published papers. It could also be the first step in understanding unpolished data.

A visually based software would help the process of introducing statistics to students. Circles are more friendly than dataframes.

Every representation shapes thought. There are limitations and distortions inherent in visually representing data. However there should be viable alternatives to the current left-brain driven paradigms.

Anki Markdown flashcards

I love Markdown and Anki.

Markdown allows me to keep "my stuff" in plain text. It has enough formatting to frictionlessly publish to Word, pdf, or the web.

Anki is my preferred digital flashcard system. Flashcards allow me to test my knowledge. Knowing what I know (and more importantly, what I don't know) is a critical step in the learning process.

I combined those two loves. You can check out my Anki Markdown flashcards here.

What is so special about the human brain? from TED.com

There are urban legends in all fields, including science. It takes bravery to challenge assumptions. To ask the simple questions with fresh eyes.

Her work is interesting but straightforward. It is simple to "count" the number of neurons. Neuroscience needs to move past counts. Move past in vitro. Move towards examining the connectome in vivo.

Only an academic neuroscientist would be shocked that human brains are not just scaled up rodent brains. Rodent models of human phenomena are powerful but past their prime. We now possess the tools (e.g., neuroimaging and cheap genetic sequencing) to more directly explore issues facing the human experience.

Evolutionary biology is a powerful heuristic for understanding diet. She missed the vital role meat plays as an efficient energy source.

If "publish or perish" is your world-view

Say yes to:

• Simple, linear, discrete, time-boxed projects within your domain
• Hierarchical / command and control supervising
• Situations where other people work on your projects with your timelines
• Thoughtful but critical feedback

Say no to:

• Complex, sprawling, open-ended projects that you know very little about
• Contributing to an “ecosystem”
• Developing talent and mentoring
• Volunteering
• Unfiltered ideas

Spend time choosing the game to play. Don’t spend time being made at the rules.

Digitizing course content

"Psychology 101" is on a trajectory to be completely digitized. For every topic, there are lecture slides (an okay option) or a video (an even better option).  It is a suboptimal use of an instructor's limited time to make and deliver introduction to psychology lectures.

However other more difficult material is on a slower digitization trajectory.

When I taught Sensation and Perception, I leveraged all digital resources I could find but still had to make and deliver lectures. Here is my small contribution to digitizing more advanced psychology course content:

You can always quit

Just because you have always done something doesn't mean you always have do it.

Every person has sunk costs. But not every person has the sunk cost fallacy.

It is always possible to change. If even if it is just your internal point of view.

A great visualization (but still a misrepresentation)

This is an example of a great visualization that serves the storytelling. However it confuses income with wealth. Income is money earned. Wealth is money owned. They are separate concepts (but often related). Wealth disparity is much higher than income disparity and is much harder to influence.

On a side note, I am glad I don't have care what the "average" American thinks. In fact it matters to very few people what the "average" American thinks. It might matter if you are a national political candidate or a marketing director for a large brand. Otherwise, it doesn't matter.

People's opinions don't matter. Just like a plane landing, it is what it is. Just because you want a plane landing to be smoother (or the economy to be more "fair"), doesn't make it that way. The forces of physics (and the market) create constraints. You can try to change those constraints but most likely you end up frustrated and tired. I suggest to option in or out. Choose to get on the plane (or engage in the market). Choose to work for an organization that has income disparity (or not).

Choosing how to show up

I choose to show up for my commitments: prepared, on time, and remain fully present.

Preparation is easy. Knowing a little about any person or topic is only a few clicks away.

On time is easy. I don't overbook myself (and leave early).

Fully present is not so easy. I'm easily distracted so I don't tempt myself. I make any possible distraction physically impossible to access.

It changes what I choose to say yes to. If it is worth my time to be there at all, it worth being the best version of myself.

A hierarchy of systems

I want the minimal maintainable systems in my life that allow me to make beautiful and/or useful things. My primary systems focus on time and project management.

Here is my timeline of system development (also rank-ordered for quality of system):

No system

A system I made up

A system collected from random parts of other systems

A complete archaic system

A complete modern system

A complete modern system modified to my personal situation

Take calendars. In the past, I did not have a calendar. I apologize to anyone who had to deal with me back then. I don’t know how I got where I wanted to be at the right time and place. I went through the hierarchy of systems, step-by-step. Right now I use Google calendar with a subcalendar for each of my different roles and responsibilities. Because it works, I spend more time using the system than working on the system.

From counts to models

Almost all data science starts with counts[1]. How many people are clicking which box? How much time are people spending on a particular page?

Only after that stage data science does gets complex (and more interesting).

There is a similar development from descriptive to inferential statistics in other sciences. Measures of central tendency (i.e., mean, median, and mode) and variance are first calculated, then models and model comparisons (e.g., regression and ANOVA) are applied.

Social media is just now entering the count stage.  Most businesses have a social media presence (binary - yes or no). They have reached a critical mass of data and are starting to organize it. The organization is counts and sums. Very few organizations are thinking about moving to the social media strategy stage (model comparisons - making choices based on data).

Many problems have followed the same pattern. Today's data scale is larger, but the analysis has the same stages. The same class of solutions can be applied at each stage.

---

1. A data scientist or statistician is usually brought in after data collection is well under way and there is an realization of its potential value. Collecting the best data in the right format probably did not happen. The classic, “I should have been at a much earlier meeting.”  ↩
   

The new gravity of San Francisco

I'm witnessing first hand San Francisco's latest tech wave.

I have been talking to many companies that are in the process of moving their offices to San Francisco proper. Other companies are apologizing for having to take shuttle buses to their palatial Silicon Valley estates.

It is interesting, possibly ironic, that Internet companies which are known for being digital are hamstrung by their physical infrastructure.

Stuart Firestein from ted.com

My thoughts:

Science is not perfect but it is the best method for improving knowledge. It makes mistakes on many levels but has built-in mechanisms to correct itself, a that process could take decades (or even centuries). Almost all other ways of improving knowing are less rigorous.

Stuart Firestein finds teaching students not exhilarating. He exemplifies one of the many fundamental flaws in R1 universities. The culture at these institutions views introduction courses (or any courses) as a necessary evil. Remember this when you are a freshman student at a big name school. These institutions are not primarily designed to give you a world-class education. It has other goals - research and grants.

It is clear from the video Stuart Firestein views himself as a “sage on stage.” Another fundamental pedagogical flaw in R1 universities (and TED).

I have taught Sensation and Perception twice. I designed the classes to show how much knowledge there is learn. I would point out where the book was wrong, incomplete, out-of-date, and suggested further research. I encouraged students to do the same. One of my goals was to inspire the desire to contribute to the body of human knowledge.

In addition to the growth of the scientific literature, there is the growth of general human knowledge. My response is not to follow but search. Even within the domains that I am “expert,” I willfully do not stay current. I spend my time solving problems. If those problems required additional “facts,” I look those up. I rather spend honing my general analytic skills than skimming RSS feeds. It does make for awkward conversions at the water cooler when someone asks me if I am familiar with a particular work or person.

Metaphors are powerful heuristics to organize thoughts. My metaphor for science is Agile Programming. I choose the next, best, completable product/problem. I ship working code/knowledge to the world in sprints. At the end of a sprint, I choose the the next, best, completable problem given how the world is and who I am at that new moment. The focus is on the doing of science for the world.

Testing serves another fundamental role in learning. Testing empirically show the limits of self knowledge. It shows where more time should be spent to continue the learning process.

PS. I think Stuart Firestein makes interesting points and does fine research. I disagree with him without being disagreeable.

The data scientist unicorn problem

I'm looking for a Data Scientist position in San Francisco Bay Area and have been on many interviews. I have found organizations are looking for too much. They want high-level capabilities in statistics, machine learning, big data, computer science, AND specific domain knowledge. Since they hiring their first (and possible only) data scientist, they are looking for those capabilities in the same person. Add to the mix non-overlapping technologies. The human resources person is looking for someone with SQL and Hadoop chops because those are the most common buzz-words. The team member wants someone with R and Unix experience because they are useful right now "in the trenches." Everyone wants years of experience in an toddler stage industry (i.e., likes to make loud noises about important it is).

Instead of trying to find an unicorn, maybe they should look for a horse with reasonable looking prosthetic horn. As Dan Savage says,  “There's no settling down without some settling for.” Find a .77 Data Scientist that you can round up to The One.

Feynman on distributed computing

@ 9:00 

That time period was an incredibly productive time in the development of computer science. Many of the ideas we take for granted about computer architecture and algorithms were first put forward and developed. The mash-up of different fields of study created fertile ground for breakthroughs. The teams were able to deliver immediate value AND develop theoretical predictions for the future.

Now could be a similar time. Computer science should look to the fields of topology, bioinformatics, and cognitive neuroscience for the inspiration of new ideas.

(Having geniuses doesn't hurt.)

The 3 Ds of skill learning

Increasingly, the interesting problems in the world require computational approaches. I don’t have a computer science background but I have in the trenches experience with computational approaches, primarily in MATLAB. However, there are hard edges to MATLAB for the problems that I think are worth solving. Python has the horsepower and the wheels to solve those problems. I applied the 3 Ds of skill practice to my quest to learn Python.

One side note about study materials. I started with free high quality resources (e.g., Google’s Python Class, Learn Python The Hardway, and Think Python) but I wanted to be more than an Advanced Beginner. I was pleased with the in-depth coverage of Learning Python and Programming Python by the fine people at O’Reilly.

Deliberate

I always had clear, concrete goals to improve my ability. At first, it was working through specific chapters or sections. This worked well for the initial introduction to the language. Later in order to find objective benchmarks, I had to rely on testing.

Testing is critical, but often times lacking in self study. It highlights which material has been mastered and which material need more attention. This learning recalibration is needed to stay in the zone of proximal development.

The best method for testing programming skills are coding interviews, and the best coding interviews can be found at the req.

Deep

My success was correlated with the depth of my studying. At a coarse level that meant no TV, IM, or Twitter. At a more subtle level, I made sure nothing was on mind other than the current learning. A trusted system was helpful.

Distributed

This classic behavioral psychology trope works every time. The same amount of practice spread out will increase retention. I would review previous material before engaging with new material. Anki is my preferred method for stacking and tracking my learning corpus.

The art of misdirection from ted.com

The above video is an engaging and intuitive introduction to attention. People are often seduced by consciousness, their own "Frank." Because the mind is the most dominant (and most frequent companion), people think that it is the most important mental process. Our brain has many mental processes, those other processes are quieter or completely silent. I am interesting in brain research, especially neural decoding, because it provides a window into those processes.

The direction and magnitude of connectomics

The connectome meme is currently the best paradigm for understanding the brain. Identifying the connections in the brain is a prelude to a far more powerful story - the direction and strength of those connections. Similar to betas in an regression analysis, identification is an important first step but the direction and magnitude of those weights is where the greatest promise lies.

The environments of innovation and productivity

It is difficult to require innovation or productivity, but they can be encouraged with the right environment. However the right environment, both physical and mental, for each one is distinctly different.

Innovation is finding a new way to "crank the widgets." It happens with happy accidents. Asking a random (seemly innocent question). Interacting with people. Realizing the link between two distant concepts. This is "heads up" work.

Productivity is "head's down" work. Humming along, mostly quiet. Clear crisp boundaries. "Cranking the widgets" is best done without interruption.

It is up to you decide what your job (at the current moment) requires. Is it being innovative or being productive? Then craft the environment that maximizes that potential. Environments, both physical and mental, are all too frequently accidents of time and space reflecting the past rather than the present.

The brave new world of molecular gastronomy

Molecular gastronomy, the science of cooking, is making exciting breakthroughs. One future direction could be combining cooking with genotyping. It is becoming increasingly cheap and easy to find people's genetic structure. Given the strong role genes play in taste and smell, I wouldn't be surprised if high-end restaurants start genotyping individual people to customize meals.

This idea was inspired by Cory Doctrow's Makers.

Future proof your learning algorithms

Andrew Ng's work in Deep Learning: Machine learning via Large-scale Brain Simulations is an inspiring direction for the field.

If you want long-term success, pick a learning algorithm that performs better with more data. That is a better choice than developing more complex models. The "long money" is on learning algorithms that best uses unlabeled data. There is far more unlabeled data than labeled data in the world.

Repost: The Future of Programming

I consistently discard old ways of thinking that no longer serve the person I want to be.

The most recent example is switching from MATLAB, to Python. MATLAB is a numerical analysis software that I badly bent into doing presentation and scripting. Python is an object object-oriented programming language designed to be flexible and scriptable. Even though I have more than 10 years of experience and all my programs in MATLAB, I switched to Python. The process was not without frustration and confusion (and cursing).

I realized that I needed to stop throwing good time and attention after bad time and attention.

The current role of a teacher

Given the world’s knowledge is available to anyone with a computer and the internet, what is the role of a teacher?

Curator
The information avalanche in the world is poorly organized and unvetted. It is the role of a teacher to collect the best, beautiful, and most useful nuggets. Carefully selecting and ordering the materials so students learn the right thing at the right time is an art.

Given most material is created by experts, it suffers from the tyranny of knowledge problem. Experts focus too much on their specific silo, thus creating knowledge for themselves or their peers.

Condenser
Books are condensers. A great book is often a lifetime of experience and insights distilled into a few hundred pages. Teachers should do the same - concentrate a wealth of material available in the world. That happens at the course and class level.

Catalyst
A teacher should be an agent to both start and spend up learning. Some students don’t realize the importance of learning some material. Other students want to learn but don’t know where to start or where the path is.

Conduit
Teachers should model the love learning and knowledge. Show it is okay to have a passion and be a little nerdy.

Critic
There is a place for gentle push back on students. Teaching (especially with older students) is often replacing misunderstandings or incorrect thinking with better alternatives. Rarely does a teacher make the first marks on a tabula rasa.

A direct assault on deeply-held beliefs is rarely effective. After rapport is established, gentle nudging is a better method to get a mind thinking in a different way.

Creator
Lastly, there is creation in teaching. It could be ground-breaking pedagogy, more often it is simply synthesizing two different concepts.

Whatever the exact nature, the world is a better and more interesting place because of teaching.

The bad science of TED

I love TED talks but some of their science is terrible.

Mark Killingsworth claims to be a scientist but does not play by the rules of science.

I am giving him a scientific yellow card at 6:06. Where are the errors bars? Additionally, the truncated range of the Y-axis is misleading.

PS Despite his misleading visualization, his study is a great example of leveraging the power of the Internet to change the nature of psychology data collection.

Deciding to directly share

I have decided to directly share more of my work.

My resume and CV are the most frequently shared parts of my work. However, they are not The Work. They are information about my work. This is similar to Google, Facebook, or Wikipedia. Google is information about the web (not the web). Facebook is information about your friends (not your friends). Wikipedia is information about the world (not the world).

These abstractions are useful but there needs to be a directness to The Work.

My computer code can be found here. My presentation slides can be found here.

A cheat sheet for R

Following up to my cheat sheet for MATLAB, I made a cheat sheet for R.

 R for the Faint Hearted 

It is a fun, quick start guide to a powerful programming language for statistics. (Yes, I used the words "fun," "quick," "programming language," and "statistics" in the same sentence).

It is a work in progress. I gave myself a constraint of 2 pages to make sure I shipped something sooner rather than later. There are many improvements to come (e.g., logical relationships, distributions, statistical tests, and plotting).

Enjoy!

The increasing need for nonlinear heuristics

Spending time in San Francisco, with its subtle but frequent earthquake reminders, makes me think of how poorly humans deal with nonlinear effects. Humans intuitively understand linear effects - a small change will result in a small effect and a large change will result in a large effect. Many change-effect relationships in nature are approximately linear. For example seasonal temperature change. In the Northern Hemisphere, temperatures typically wax during the spring and summer and wane during the fall and winter as a result of the Earth's axis tilt.

Nonlinear effects take us by surprise. Nonlinear natural events are more difficult to understand. Insert your most salient natural disaster.  If change-effect relationships deviants from linear, we fail to understand (or predict) the results.

In contrast to the natural world, the human-driven world is becoming increasing nonlinear because of interconnections. That means the total number and impact of the nonlinear events are increasing. To deal with this change in the world, new heuristics are needed. Our "natural" (linear) heuristics are no longer sufficient.

Start with The What; Then comes The Why

"The what" is the surface pheonmeom. How it appears. The data.

"The why" is the underlying mechanisms. What causes it. The reason.

Another life lesson gleamed from teaching biological psychology.

Algorithms & heuristics

I collect algorithms as I make my journey through a career as a Computational Cognitive Neuroscientist /Computer Programmer/Quantitive Analyst/Data Scientist. An algorithm is a rigorous method to solve a problem. The best ones are simple and elegant. My all-time favorite is the linear algebra behind the General Linear Model (GLM), which includes ANOVA and regression. A recent project was helped with modular arithmetic. It was a computational hot knife through programming butter.

I also collect heuristics, less rigorous methods to solve problems. The best ones are simple and elegant. My all time favorite heuristic is “past, present, and future.”“ Describe the past. Capture the present as it is (to the best of your ability). Envision a future. Thanks to Charlie Munger’s tome I started to use ”Always Invert." If you only have access to the properties that don’t work, you can invert them to find the properties that do work. For example if a project failed because of scope creep, try to limit the next project to the smallest meaningful, shippable unit.

Algorithms and heuristics are “force multipliers.” They can solve problems more quickly with less effort. They are tricks, ways to make the complex simpler. They free limited cognitive resources so the same problem doesn’t have to be solved twice.

Algorithms are powerful, but narrow and brittle. Heuristics are weaker, but wider and more robust. The art is knowing whether the problem should be solved with a algorithm or heuristic (or even an unique solution).

Flipping the yoga studio

There is a spectrum of yoga instruction from rigid (e.g., Bikram) to freestyle (e.g., Steve Ross). The evil genius of rigid instruction is “optimization,” always the same, perfectly constructed sequence.

If it is always same, what is the instructor’s role? They typically don’t demonstrate the poses; They usually just give the same rapid fire instructions every class.

Why not record the instructions and play them back? This would free the instructors to give individual feedback to the students.

A similar approach works in academic classrooms. No one should deliver another intro psychology lecture. They are all available online. It is a better use of class time to replace lecturing with feedback-dependent activities, for example presentations, projects, or testing (both formal and informal).

Returning to San Francisco

I’m excited to return to San Francisco. The art. The culture. The fog. The hipsters.

I spent 4 years, ’97-’01, in San Francisco while earning my undergraduate degree. I was focused on my studies and missed out on the dot-com boom (and bust). I moved down the California coast to follow my education dreams. During that time, my focus was research, and I missed out on Web 2.0.

Now is time of another internet revolution, aka Web 3.0. This one will be focused on the copious amount of data that is being generated. No one knows what it means or what can be done. It a great time to be "in the game."

All of my professional interests have been pointing to the field of data science, the union of statistics, programming, and communication. Now is the time to make the leap. San Franscisco is the epicenter of a future world that I want to contribute to.

Repost: The surprising seeds of a big-data revolution in healthcare

I am also passionate about solving meaningful problems with data. It wakes me up early and keeps me up late. It is "the why."

"The how" is becoming increasing easier. The tools of automatization are accessible to everyone (even hackers like me). I'm exploring how to automatize the data collection process for psychology experiments. My playground is Online Psychology Experiments.

Markdown for teachers

Markdown is a light-weight HTML syntax, primary used by web writers. It was my “killer app” during my last semester of teaching.

There are 3 places my classroom writing appears: handouts, slides and web. I use a Mac so my handouts are in Pages, my slides are in Keynote, and my webpages are in Canvas (the university’s chosen CMS). If I write all my materials in plain text with Markdown formatting, it trivial to translate into any or all of those mediums. Using Marked, I convert my plain text to rich-text formatting (Pages and Keynote) and HTML (Canvas). The same “stuff” appears in each place and is appropriately formated for each location.

Additionally, the content files are small, nonproprietary, and come with me everywhere.

Markdown is well worth the minimal time and effort it takes to learn.

PS The best way to learn Markdown.

Missing a mentoring opportunity

I was recently at a small event during which well-established faculty gave talks (over an hour a piece) to graduate students, post-doctoral fellows, and junior faculty.

Every faculty member missed the opportunity to be a mentor to these young scientists (including me). They spent the entire time “down in the weeds” of their personal science. The content of talks was engaging but they missed the unique opportunity of the room.

They could have offered a peak behind the curtain. They could have shared the patterns and anti-patterns of academic research. They could have shared how they choose their research questions or their failures (everyone in science has ’em). They could have been vulnerable. Instead, they choose to be a “sage on stage” and share only their successes. If I wanted just the information, I could read their published papers.

It is the difference between screening a film and giving the rules of storytelling.

PS In contrast, E.O. Wilson’s book “Letters to a Young Scientist” is an example of remote mentorship. He was able to find the themes from his personal journey applicable to others.

A biological psychology teaching taxonomy

Coming off a back-to-back stint of teaching Sensation and Perception, I have developed several heuristics to aid my pedagogy. One of the most powerful is a taxonomy I have developed for biological psychology instruction.

I start by outlining a conceptual understanding of the material, providing an overview and context. The students should understand the big-picture and how the topic applies to other topics and their personal lives. For example, the retina of the eye is one type of sense organ that transduces environmental energy into neural signal. The retina processes light rapidly and with great acuity.

After a conceptual overview, I outline the structure. Being able to identify and label structure elements builds a visual scaffold for latter learning. Continuing with the visual system example, the retina is a complex structure with several layers each with specific shaped cells.

Structure seamless leads into function, how the particular system works. This is where I spend the majority of my teaching time. I hope to leave the students with a deep appreciation and understanding of how  biological systems function. In the retina, the fovea of the retina has greater acuity (function) due to cone density (structure).

If time allows, I will go into the details. Often times, professors move to this level to quickly. This is where things are most interesting but also the easiest place to lose students. An example of detail in the retina is the enzyme cascade in the photoreceptors which changes neurotransmitter release. If a student does not have an overall understanding of the structure and function of eye, then the role of rhodopsin will lack "stickiness."

The last layer is the numbers. In undergraduate classes, I rarely emphasize numbers. Some students maybe able to memorize them for an exam. A majority of the numbers from a class will not be remembered in the long-term. Stories and concepts will endure. For example most humans have 3 different cones, each one has peak sensitivity at a specific electromagnetic wavelength. It is far easier to remember there are Short, Medium, and Long cones than remembering 564–580 nm, 534–545 nm, and 420–440 nm cones.

This taxonomic gives general framework to teach any biological psychology concept. It is also a fractal pattern, from nervous system level down to individual cells or from introductory to advanced topics.

Who The Brain Initiative is picking

The Brain Initiative has been the talk of my “tribe” (i.e., cognitive neuroscience researchers). Right now the field has the tools and technology to make rapid breakthroughs. A large influx of funds has the potential to catalyze those breakthroughs. 

However the funds are primarily targeted at large organizations, which makes less sense in today’s world where resources are increasingly available to smaller organizations. Smaller organization are more likely to push the edges. Smaller organizations are also more likely to fail (hand-in-hand with pushing the edges is the potential for failure). The problem is failure looks bad for funding. The point of funding should not be to pick guaranteed “winners.” It should be the chance to great work.

The Brain Initiative should think more Silicon Valley, less Detroit.

A perception professor visits The Exploratorium

I just attended the newly renovated Exploratorium in San Francisco. It is Disneyland for STEM geeks.

My favorite exhibit was an analog square root calculator. You place a ball at a numbered location on a ramp. The ball then rolls down the ramp and flies through the air. It hits a metal rod associated with the square root of the number on the ramp (e.g., a ball at the #16 ramp location would hit the #4 rod). The Exploratorium sets it up as an experiment to discover this relationship. It then offers a short story, a simple explanation, and a long story, a more complex explanation including equations. This is science education at its finest; Experiential learning with the appropriate explanation of mechanisms.

I was most interested in the perception and cognition exhibits. They covered the gambit from low-level sensation to social group dynamics. One of the highlights are different shaped bellows that mimic human vowel sounds. This is an idea I am bringing back to the classroom to better explain the uniqueness of human speech.

Despite the quantity and quality of exhibits, I felt The Exploratorium was lacking the long story in the perception exhibits. My appreciation for the wonders of the human nervous system increases by knowing more about the underlying mechanisms. In particular, there are many equations that can capture important properties of human perception but they were incomplete (or missing entirely).

The current Exploratorium overloaded me with ideas. I am already looking forward to my next “field trip” to see how it grows into its new space.

If I only know how to become a fellow …

Researching what we all ready know

More people research vision than all the other senses combined. That is ironic since we know the most about vision.

It is similar to business. People are more likely to start stores they already see, more willing to be the second coffee roaster than the first coffee roaster in town.

It is scary to deviate from the known, therefore the common. However in the unknown lies the most opportunity.

Evaluations from my first course

Teacher feedback is a hot topic in the field of education. I realize its value to both my students and me so I have built-in feedback mechanisms in all the courses I teach so I can make real-time corrections. They range from in-class polling to on-line surveys. I recently received the traditional end-of-term (aka, too-late-to-make-changes) evaluations from Sensation and Perception at Catholic University of America for the Fall of 2012.

Given it was my first course, there were rough parts. I over-estimated the knowledge students had coming in. I also over-estimated my ability to quickly and clearly cover the technical material. I need to slow down and elucidate. However, those peccadilloes were dwarfed by my passion and preparation. Care is my trump card.

Don't take my word for it. I will let the students speak for themselves: The evaluations from my course, including the negative ones. (The last one is open for interpretation.)

They show my potential for becoming a college professor. Now I am doing the hard work of actualizing that potential.

Picking projects

Given: All project resources are finite.

Do you choose the largest research project you have resources for? Call in all your favors. Push every piece of equipment and person to the limit.

Or do you choose the minimum project that you have resources for? Make something meaningful but leave something "in the tank" for the next project.

There are infinite places to choose along that continuum. I choose my place by remembering: Individual projects fail (regardless of resources), and I plan to have a long career.

Happy commencement?

Now is the time of year for speeches and platitudes. I have plenty of both, but I usually keep my mouth shut.

It is also the time of year for choices and advice. When students ask for advice, my first instinct is to give advice to myself in the past. My second instinct is to give advice to myself in the present. I stop both of those disingenuous instincts.

Instead, I ask questions. My favorite questions is "What would an ideal day look like for you in 4 years?"

Most people don't have an answer.

That seemly simple question is at the heart of life. A good life is a series of good days.

Tim Cannon, biohacker, talking to my Perception class

Tim Cannon of Grindhouse Wetwares recently spoke to my Perception class (PSYC 310) at The University of Maryland about his (literally) cutting-edge work.

No one picked him to do that work. His doesn't have a grant or an academic position. He decided to organize like-minded individuals and leverage current technologies to make the world a more interesting place.

Are your methods as good as your theories?

Humans are meaning makers. We find patterns and stories wherever we look, from a side-wise glance across a bus to who-sits-next-to-whom at a table.

That meaning making is a double edge sword in science. It helps draw the novel connections of scientific breakthroughs but also allows for wishful thinking about muddled data. The scientific method, research methods, and statistics are "catch trials" for that process. They help separate signal from noise and prevent scientists from being too human.

The scientific method, research methods, and statistics are considered the basic elements of research. They are taught first, followed by domain specific knowledge. Those basics are rarely revisited. The opposite should be true: Those basics should not only be revisited but continually honed.

Improvements in domain specific knowledge need to be echoed with improvements in the methods that create that knowledge.

Congratulations to my research assistants

Veronica Kwok and Hamza Raja capped off a successful semester of research with a poster about our decision-making project. They pulled a double-header, presenting at both the Psychology Department Research Fair and campus-wide Undergraduate Research Fair.

Veronica is starting her graduate studies in the fall at The University of Maryland. Hamza will be continuing his undergraduate studies.

pdf

The problem of meaning in the connectome

Connectomics has the potential to move the field of neuroscience forward. It can enliven the static perfection of neuroanatomy and connect the isolated islands of activations from fMRI.

It faces the same challenge that all big data faces: There is no taste. The connectome needs curation. A human to decide which of the several trillion synaptic connections in the brain are important. Computers are superb at gathering the raw materials but are woefully poor at synthesis.

As data grows bigger so does the need for human discernment.

Neuroimaging project management insight

Errors scale faster by size. Small projects are more often on-time and if late, they are late by smaller amounts compared to larger projects. Think about trip length, compare going to the corner store to going to another continent.

Neuroimaging projects are inherently larger than cognitive psychology projects. Neuroimaging requires more technology, more data, more statistics, and more team members than cognitive psychology. That increased size increases the chances of a project being delayed, possibly in non-linear ways.

It behooves someone to be a varsity-level project manager in cognitive psychology before moving into neuroimaging. "Jumping levels" is a good way to stop moving.

Please do not report the average of a bimodal distribution

The authors of Power failure: why small sample size undermines the reliability of neuroscience state, “Our results indicate that the average statistical power of studies in the field of neuroscience is probably no more than between ~8% and ~31% …”

That is an unacceptable statement[1] given the bimodal distribution of their data. Average (i.e., mean) is a measure of central tendency. Bimodal distributions do not have a central tendency (by definition). That statement reduces the richness of their data to the point of distortion.

---

1. It is irrelevant they report a range instead of a point estimate.  ↩
   

Defining a postdoc's job

A postdoctoral fellow's job is research. Research has many shades. If the goal is to publish, the projects should be entirely within the current knowledge base and skill set. If the goal is to develop research skills, the projects should just out of reach with ample guidance. If the goal is to mentor others on their research path, the projects should be well within the mentor's domain.

There are distinct and optimal work environments for each of those jobs. If you have all the knowledge and skills, the work is best done alone and uninterruptible (i.e., the novelist in a secluded cabin). Almost all interruptions will slow down the process. This is closed (and sound-proof) door work. If the research requires growth and collaboration, then access is needed. Digital access is fine but the exchanging of air molecules is far better. This is door opening "out" work. On the other hand, mentorship requires a door opening "in."

The typical office or research lab is only ideal for the last situation. The typical research environment does not have closed doors or the presence of the principal investigator. It does have frequent interruptions, often about "cake in the break room" or one person asking another person to be a human Stack Overflow. Those interruptions slow down the first two types of research. However if the goal is mentorship then those are not interruptions, they are teachable moments to make a human connection.

2013 Cognitive Neuroscience Society Annual Meeting Highlights

William Newsome Keynote Address

William Newsome does ground-breaking single-unit recordings of awake, behaving primates, the grist for my computational modeling mill. He shared new recoding data and his modeling of the data. The subject matter is engaging, gating of sensory information for decision-making, but the modeling is a game changer. It is the work of a post David Sussillo. The modeling is a premier example of dynamical neuroscience. First, he identified the relevant components to the signal. Second, he tracked how the relevant contribution of each of the components change over time. That might appear to be the next logical step but it is an additional dimension to the data, creating a wealth of information to explore. It was disappointing there was no question and answer period after the talk. A well-moderated Q&A would have added to both the science and the entertainment of already exceptional talk.

Analyzing Patterns Of Brain Activity To Understand Human Vision And Cognition Mini-Symposium

Frank Tong and John Serences were the warm-up for the main show of Jack Gallant. He researches the encoding and decoding of movies during fMRI. His work is much more than a “mind reading” pallor trick. He is able to identify the distributed semantic meaning network in the brain for percepts. In addition to what he does, I am also inspired by how he does it. He leverages cutting-edge web technology (i.e., HTML5) to facilitate the open exploration of his data. You can confirm (or disconfirm) his conclusions and explore your hypotheses All of this and more can be found on his lab website.

Posters

There were many interesting posters, here are my favorite two posters:

Lately I have been learning Python to tackle “big data” challenges. It was nice to see an application of this idea in Building Connectomes From The Cocomac Database Using Cocotool. The researchers data mined a macaque anatomical database with CoCoTools for Python. I have been playing with the code, it is simple, clean, and fast. I look forward to applying similar methods to human anatomical data.

Sebastein Hélie, a postdoctoral fellow while I was a graduate student in the Laboratory for Computational Cognitive Neuroscience, presented A Computational Cognitive Neuroscience Model Of Criterion Learning In Rule-guided Behavior. It included a new model for hebbian learning at inhibitory/GABA synapses which captures the properties of criterion selection during rule learning. The next challenge will be to translate the model to dopamine dependent learning.

General

This conference (and San Francisco) were a welcome working vacation. I have been doing mostly the “head’s down” work of research and teaching. It was nice to switch to the “head’s up” work of watching lectures, interacting with poster presenters, and having spirited conversations in the halls. I am reenergized to do the challenging work of cutting-edge science.

It was also a pleasure to catch up with old friends and colleagues and start new friendships and collaborations.

The current and future me

The current me always predicts the future me will have the same thought process as the current me. I will be in the same state of mind (with the same amazing insights).

However, my thought process is constantly changing. That churning is a source of endless creativity but also makes it hard to find my car keys if I don't put them in the same place every time.

The habit of writing things down communicates between these temporally-separated selfs. I frequently write short notes, letters mailed to my future self. This blog is an example of those letters.

Hanging out San Francisco for CNS Conference

I am in San Francisco for Cognitive Neuroscience Society Annual Conference from now until April 16.

If anyone wants to grab coffee, send me an email bspiering@gmail.com.

Repost: 4 pillars of college success in science

I am also a geeky kid-at-heart who wants to make a difference in the world. My mission is making cognitive neuroscience assessable to everyone.

When I teach, I have high expections and expect hard work, and I want everyone to succeed.

This is personally relevant since I also teach at the college level in Maryland. I personally know the issues he faces. It requires taking a diverse collection of students from where they are currently to where they want to be.

Minimizing static knowledge thinking with a simple heuristic

Glossing over controversies and unresolved issues is a necessary evil in undergraduate courses. There is often not time to provide the context to appreciate subject area nuances.

As a result, sometimes undergraduates perceive a subject area as more static than it truly is. They falsely believe al the interesting and assessable questions have all been answered.

I try to avoid that cognitive trap by applying the "Past, Present, and Future" heuristic. For any given issue, I briefly cover past perspectives. I spend a majority of the time on the present, bookended with a discussion of possible future directions. This encourages the students to realize that knowledge is not static and they have the potential to contribute.

Introducing Google Nose

Vision dominates the human experience but that is no reason to exclude the other senses. Apple with iOS has realized the value of touch. Until now the chemical senses have been underserved in the digital marketplace. I plan to use Google Nose in my current Perception class to help explain the complexity of olfactory sensation and perception to my "digital native" students.

It is amazing to witness the cutting edge of sensation and perception.
(Some might say over the edge).

Serving science through null results

There are two reasons for null results. First, there is a true effect that was not observed. Second, there is  no true effect.

If you believe the latter is true, science is best served by publishing the nulls results. Given that "publish" is now a button on everyone's computer, it easier than it has ever been to put a true null result out in the world. Holding back is robbing the world.

Academic fear

Recently I attended a conference on learning. My primary interest was the talk on molecular signatures of synaptic plasticity. (Keep an eye on Daniel Pak, he does brillant work.) My attention was also grabbed by the last talk on Massive Open Online Courses (MOOCs). In a fascinating talk, Randy Bass laid out the current state of online learning. The take home message was: The world of teaching is changing. It is unclear what higher education will look like in the future, but it will be different.

A distinguished academic spoke-up at the end with palpable fear in her voice. She was afraid and resentful that the tools academics are developing for online teaching will later destroy the institutions that created them, similar to when the Internet is used to attack the Defense Department.

I felt compassion for her. Her love of teaching was clear.

But the technology don't care.

The world is going to change. You can get upset but it is still going to change.

Instead, I am choosing to get on the ride.

Research is 90% listening

"Singing is 90% listening" is an an old quote. You must pay close attention to the world to a make meaningful contribution. Research is similar. Deeply understanding research is a challenge. It is fun but still a challenge. People have pet theories and personal biases. Those dominant over actual content. What "is" there takes a back seat to what "should be" there.

Brain Awareness Week

This week is Brain Awareness Week. We now have the tools widely available to look at brain function, even if you are an amateur scientist (here or here). Given the tools are waiting, the world could use more people trying to solve interesting problems.

An observation about predictions

Most predictions reveal more about the predictor than the subject.

How fragile projects make an antifragile career

Nassim Taleb argues that an antifragile whole is best composed of fragile parts.

I knew and applied this concept implicitly before Nassim stated it explicitly. At any given time, I have many separate research projects. Each one of those individual projects seeks to answer a specific scientific question. The questions are so specific the projects become fragile. A small deviation  (e.g., conceptual flaw, lost data, or a misinterpretation by a reviewer) leads to the "death" of a project.

However, the death of a single project strengthens the others. Through a rigorous post-mortem process, I find what does not work and apply that knowledge to the rest. A plethora of micro-failures, though painful at the time, creates an antifragile career.

Publication: Corticostriatal contributions to musical expectancy perception

I am proud to present my latest publication - "Corticostriatal contributions to musical expectancy perception." This is my first neuroimaging publication! Additionally, I am a rabid music fan, and it was a joy to study a topic that I have a strong personal interest in.

Abstract: This study investigates the functional neuroanatomy of harmonic music perception with functional magnetic resonance imaging (fMRI). We presented short pieces of Western classical music to nonmusicians. The ending of each piece was systematically manipulated in the following four ways: Standard Cadence (expected resolution), Deceptive Cadence (moderate deviation from expectation), Modulated Cadence (strong deviation from expectation but remaining within the harmonic structure of Western tonal music), and Atonal Cadence (strongest deviation from expectation by leaving the harmonic structure of Western tonal music). Music compared with baseline broadly recruited regions of the bilateral superior temporal gyrus (STG) and the right inferior frontal gyrus (IFG). Parametric regressors scaled to the degree of deviation from harmonic expectancy identified regions sensitive to expectancy violation. Areas within the BG were significantly modulated by expectancy violation, indicating a previously unappreciated role in harmonic processing. Expectancy violation also recruited bilateral cortical regions in the IFG and anterior STG, previously associated with syntactic processing in other domains. The posterior STG was not significantly modulated by expectancy. Granger causality mapping found functional connectivity between IFG, anterior STG, posterior STG, and the BG during music perception. Our results imply the IFG, anterior STG, and the BG are recruited for higher-order harmonic processing, whereas the posterior STG is recruited for basic pitch and melodic processing.

You can find out more here.

Repost: A monkey that controls a robot with its thoughts. No, really.

Here comes the Technological Singularity!

This video is an union of my professional interests (neural decoding and comparative psychology) and personal interests (biohacking and cutting-edge technology).

From my limited experience with primates (all species), they will not everything you want them to do, but what they will do, they will do for juice.

"Other practices" always precede "best practices"

There are best practices in a given field. Before best practices can be established, other practices must be acknowledged.

People in science tend to "roll their own." They create their scheduling or project management systems. That is a reasonable choice if it is truly a choice. More often, people are proudly ignorant of what others are doing.

"Fundamentals of MATLAB for Neuroscience Research" talk

I have been using MATLAB in cognitive neuroscience research for over 10 years. However as the computer adage goes, "It is not the number of years but the number of hours." I use the program almost daily.

Recently, I gave a "Fundamentals of MATLAB for Neuroscience Research" talk. It was slightly odd talking about a topic that is often conceptualized as solely a doing activity. Despite that, I feel starting with a strong "why" helps to motivate the "how." The talk was followed with a Q&A period and a brief hands-on programming activity.

The slides can be found here.

I gave out the 2 following handouts: a cheat sheet and a style guide.

My dream job

My dream job is asking and answering interesting questions. I find the interesting questions through the "heads's up" work of reading and talking. External stimulation makes this work better and faster. I try to answer interesting questions through the "head's down" work of synthesizing and researching. External stimulation slows the down that work. The tension of those magnet poles energizes my day.

Undergraduate Awards in Auditory Cognitive Neuroscience (Summer 2013) at McGill University

Applications are sought for Summer 2013 (16-week) undergraduate awards offered by the Auditory Cognitive Neuroscience (ACN) Training Network (NSERC-CREATE), to undertake research in our top ACN labs at McGill University, McMaster University, University of Montreal, the Rotman Research Institute and the Montreal Neurological Institute.

Positions begin May 2013. Citizens of all nationalities are eligible to apply. The deadline for applications is March 1, 2013.

Full application and eligibility details are available here.

Repost: Hey science teachers -- make it fun

I'm trying my best to make my Sensation and Perception class fun (and teach the information at the right level).

On choosing the next research project

How do you decide your next research project?

It safest to look back at previous work, conduct a logical extension of a published study or permutate common parameters. Looking backward leads to step-by-step movement. There is a small chance of complete failure but an equally small chance at innovation.

It more risky to look ahead and choose a research project based its potential to make a contribution. Looking forward leads to moving in leaps and bounds. This is an increased chance of failure, but also the opportunity for massive inovation.

Sample "blended" lecture slides from my course

I am taking a blended (or hybrid) approach by combing on-line resources with in-person lectures for my "Perception" course at The University of Maryland's Department of Psychology.

Syllabus

Sample Lecture Slides

The connectome & the connection economy

The connectome zeitgeist is rapid moving through the neuroscience community. Like the uncertainty principle or Schrödinger's cat, this scientific idea is a reflection of the larger culture.

The connections neurons (or people) make are often more important than their individual functions.

First day of class

I was honored with an invitation to teach "Perception" at the The University of Maryland for Spring Semester 2013. Day was the first day of class.

One aspect of teaching that I love is the creation of a mirco-world. It is similar in computer programming. Within a general framework, there is ample freedom to create something special. That freedom, both in programming and teaching, is often under-realized. I plan to use more of that freedom within my current class.

If you are interested in the map for the territory I have created, you can find the syllabus here.

Bayesians are still missing a big win

Nate Silver's The Signal and The Noise argues for a Bayesian approach (i.e., thinking probabilistically, making explicit predictions, and gathering "Big Data") to make better predictions. It is limited by the fundamental flaw in the Bayesian paradigm - casual mechanisms are given no weight.

Understanding specific causal mechanisms is one of the biggest "wins" for predication. It is not always possible, but the Bayesian paradigm does not value even trying.

Global vs. local science

The world's collective knowledge is instantaneously available. It is just sitting there waiting for me.

However, the world's collective knowledge is wide but shallow compared to the narrow but deep impact that specific mentors had my intellectual development.

I choose every day which type of impact I make.

Inspired by this post.

Answering the "what" and the "how" in science

Empirical investigation of surface phenomenon is a common departure point for science. The "What is going?" question is best answered at the bench or in the field.

On the often-neglected flipside is the search for the mechanism underlying that surface phenomenon. The "Why does that happen?" question is best answered at the whiteboard, fueled by coffee, or on the back of an envelope, fueled by other adult beverages.

Both are the work of science. The former is the long work. The later is the hard work.