Bill Gates has just declared this week as Mosquito Week, pointing out that the mosquito is the cause of the most human deaths per year of any creature on earth, even exceeding us. In our lab, in collaboration with Jeff Riffell and Michael Dickinson in the Department of Biology, we are working to understand the neural and computational mechanisms that enable mosquitoes to find their prey using joint signals from the heat and carbon dioxide that we exude. This problem is an especially fun one for theorists: because of turbulent transport, the source signals are complex and time-varying, with interesting statistical properties; and the two signals interact in nontrivial ways, as Leslie Vosshall’s lab has recently shown. Here’s the beautiful graphic from Bill’s blog posting.
Former lab member Mike Famulare was recently interviewed by Seattle Times’ NWJobs. After graduating with a PhD in physics in 2012, Mike joined Intellectual Ventures’ research effort on modeling disease propagation and vaccine efficacy– an excellent use of his stochastic-process chops.
The edition of Current Opinions in Neurobiology on Theoretical and Computational Neuroscience that Haim Sompolinsky and I co-edited has just been published. I’m excited about the diverse and fascinating articles our authors produced, giving a snapshot of the field and an interesting view of where we are headed. Both Haim and I wove our own impressions of what we took from this collective wisdom. Check it out.
The tech world is also in the grip of a justifiably renewed excitement about artificial intelligence. It was intriguing to see so many students from tech in our Computational Neuroscience MOOC (mentioned in this article), which ended its second session last month. I’m optimistic that this increased interest and communication will help to drive discovery in fundamental neuroscience.
Many people have been in touch with questions about how to pursue further education or careers in computational neuroscience. With the BRAIN initiative in the news and a number of companies launching into neural technologies, this is certainly an area of current opportunity. I wanted to give some perspective on possible trajectories, as the field is very diverse, spanning academic study in quantitative approaches to systems neurophysiology to robotics and industrial engineering.
It is not necessary to join a graduate program in computational neuroscience. Indeed, I am not strongly in favor of such programs and here at UW we have chosen not to go that route: I believe that computational neuroscience is the quintessential interdisciplinary program, and it is important to master at least one discipline. So I tend to favor basing oneself in a home department that provides the core skill set and knowledge base you seek: neurobiology, computer science, physics, statistics, electrical engineering, bioengineering or mathematics, but doing so in a place with opportunity and proven track record of cross-departmental collaborations and openness to interdisciplinary thesis projects. That way, you will ensure that your core coursework is a strong preparation for one academic discipline, but that you have the chance to take a variety of electives that allow you to get a broad training in other areas, and that your interdisciplinary research will be fairly and helpfully guided and evaluated.
Where? Top-tier examples of institutions in the US that provide such an environment (in no particular order) include the University of Washington (naturally!), Berkeley, UCSD, Columbia, NYU, Carnegie-Mellon/Pitt, Princeton, MIT, the University of Chicago, U. Penn, UT Austin, Northwestern. Other notable options include Duke, UT Houston, UC Davis, Brandeis, Yale, Stanford, Caltech and USC, which has an impressive retinal prosthetics group. It is important to be aware that entry into graduate programs at all of these institutions is highly competitive. As an example, the University of Washington’s Neurobiology and Behavior graduate program received nearly 300 applications this year for an anticipated class of 10, and only around 30 students from an excellent applicant pool were interviewed. So it is important not to fixate only on top programs, but to look carefully at some other places where you can find at least two or three researchers who are working in areas that interest you. You might be able to develop a novel collaboration between faculty who have not worked together before. Very few institutions have a “computational neuroscience” program. You should look to see which graduate programs (eg. Neurobiology and Behavior, Computer Science and Engineering, Applied Math) host the faculty whose work interests you, and where you will be able to do the coursework and research you have in mind: there may be multiple options. Also be careful to tailor your choice of institution and program to the direction that you are motivated by, whether that be delving into the basic mathematics of chaotic neural networks, probing decision-making behavior and/or circuitry in primates or rodents, working closely with clinical applications or developing devices. All of these require rather distinct training.
Internationally, there are centers of excellence in many countries, and this area has recently received an enormous stimulus in Europe in the form of support for Henry Markram’s brain simulation consortium, centered in Switzerland. In terms of clusters of individual labs doing high quality systems neuroscience with theory components, Germany, Great Britain and Israel have perhaps the most extensive offerings. Germany boasts the network of Bernstein Centers for Computational Neuroscience and many Max-Planck centers have been trailblazers in systems neuroscience, particularly in invertebrates (e.g. Tuebingen). Britain has a long-standing tradition of quantitative approaches to biology and the support of the Wellcome Trust for systems neuroscience; there are particularly vibrant groups at UCL (including the Gatsby Institute), Cambridge, Oxford and Edinburgh. Israel has the Weizmann Institute and Hebrew University with very strong programs, and good ones also at Ben Gurion and the Technion; the Weizmann accepts foreign students although a better all-round option for a stint in Israel may be for a postdoc. I’ll also randomly identify a few other places that I know have interactive theory/experimental groups: Ecole Normale Superieure in Paris, ETH in Zurich, the Champalimaud center in Lisbon. But again, there are many places where high quality work is going on at lower concentrations that would be fine choices.
Getting into graduate school. The admissions process focuses less on the specifics of your undergraduate training, reflecting the diverse backgrounds that are appropriate for neuroscience, but on high academic performance and interesting and productive research or work experience. Research experiences tend to be favored over work experience with the assumption that these reveal more of your independence and creativity, so if you are applying out of a work situation, your statement and letters should emphasize and give specific examples of these qualities.
Courses and Summer Schools
Of course there is our Coursera course, which we are likely to run regularly. There are also some excellent summer schools which can afford a quick boot-up in the area. One of my favorites (as I attended it and also directed it for five years) is the MBL course, Methods in Computational Neuroscience. This is a very selective course aimed at graduate students and postdocs which takes one through the mathematical methods that underlie many contemporary approaches to problems in systems neuroscience. Other options are the Okinawa course run through OIST, and Cold Spring Harbor Labs has initiated what seems to be an annual course in China. CHSL has a highly regarded Computer Vision course that counts as alumni and faculty many of the top researchers in vision neuroscience. There are also several European courses including through the Bernstein Center: watch the compneuro mailing list for opportunities.
A few companies are working on fundamental neural research with the hopes that this will lead to useful devices. One such company is Brain Corporation, in San Diego, which is building a model of the retina with a long-term view to modeling the entire brain. The Allen Institute for Brain Science, in Seattle, has also embarked on a long-term project to solve the problem of coding in the visual system using a combination of high-throughput electrophysiology, anatomy and modeling. The Redwood Center for Neuroscience, in Berkeley, was also initiated as an academic/industrial collaboration to transfer fundamental brain algorithms to device design. IBM Alameda has an ongoing project to replicate cognition in a brain model.
A much larger number of companies are entering the field of neural prosthetics and brain-computer interfaces. One way to locate such opportunities is to look for university-based centers which have industrial affiliations and partnerships; their websites can direct you to those companies. One example is UW’s Center for Sensorimotor Neural Engineering, an NSF-funded Engineering Research Center. There are perhaps a dozen comparable centers across the US. These companies may offer internships and other ways to gain experience and refine your understanding of the qualifications you require.
Thanks to information provided by our Coursera students, edited to add the following:
“UK tuition fees for postgraduate courses (if you fund them yourself) are usually about £9,000-£11,000 a year for EU and about twice that for non-EU students, but there is a lot of variability in the cost, more so than for undergraduate courses. When you look for funding, then the University is one option, but depending on where you are from, there may be additional opportunities for sponsorship.”
Please feel free to add links and information about other programs in the comments!
In late September, our computational neuroscience program held its annual Computational Neuroscience Connection, a one-day meeting on the UW campus for groups involved in this area both at UW and beyond. Sponsored by the Center for Sensorimotor Neural Engineering, the meeting’s main goals are to provide a forum to meet other groups and present and learn about new research.
This year’s event had many highlights, including excellent talks from our students and postdocs representing a broad range of research from theoretical network modeling to neural control of flight to cochlear implant design. Clay Reid of the Allen Institute gave a brief overview of the Mindscape project and went into detail on his recent studies examining the detailed connectivity patterns of specific cell types in cortex. New faculty member Emily Fox discussed her advances in statistical modeling of EEG data. Cris Niell, of U. Oregon, described LGN responses in awake behaving rodents and noted a surprising response type, a neuron that is completely silenced by any contrast in its visual field. Perhaps most dramatically, Emo Todorov closed the day by showing astonishing examples where principles of optimal motor control combined with constraints on surface contacts can predict and model remarkably lifelike human motor behavior.
Earlier this year Bill Bialek organized a little symposium at the Initiative for Theoretical Science at the CUNY Graduate Center, on Adaptation, Inference and Natural Sensory Inputs. Speaker Wei Ji Ma began with a wonderful quote from Al Hazen (965-1040) that captures the essence of what many of us are trying to understand:
“Not everything that is perceived by sight is perceived through brute sensation: instead many visible characteristics will be perceived through judgment, in conjunction with the sensation of the form that is seen.”
I had the pleasure of participating in the Bernstein Center for Computational Neuroscience’s 10th Summer Course in Goettingen. These annual courses are, remarkably, organized entirely by students. They feature five speakers in the course of a week, and have a very productive format. After 3 hours of tutorial lectures in the morning, students break into small groups to prepare and then present a list of papers provided by the speaker, consolidating the material, getting experience in rapid information digestion and honing presentation skills. Surya Ganguli gave a very elegant presentation on compressed sensing and its applications in several areas of systems biology; Susanne Schreiber discussed subthreshold resonance in single neurons and a possible connection with grid cell responses; and Matthias Bethge showed mixture-of-Gaussian natural scenes models derived from principles of redundancy reduction that are capable of state-of-the-art image compression and texture synthesis. Thanks to our excellent hosts, David, Agostina and Max. As organizers, these three gave great training to the students with regard to their paper summaries and presentation skills.
The Max-Planck Institute for Self-Organization and Dynamics in Goettingen sits in the forested hills above the city and is a very attractive working environment, with a spectacular collection of theorists and experimentalists in close proximity and a bright and open design with appealing areas for discussions. I very much enjoyed discussions with several local scientists, especially Andreas Neef, Fred Wolf and Ahmed al Hady (one of our previous MCN students!) about their beautiful optogenetically-activated networks cultured on multielectrode arrays. This arrangement allows them to drive and record extracellularly from cells over many days. In the future, this preparation might be used to explore Fred Wolf’s group’s very interesting theoretical findings about the dependence of dynamics of balanced networks on intrinsic neuronal properties.
Goettingen itself, a delightful walled city full of well-preserved half-timbered houses, is an intellectual capital, home to many great scholars and thinkers since the founding of the university in 1734. Plaques mark the former dwelling places of hundreds of these scholars. The city is perhaps proudest of Gauss, whose life we traced in a fascinating walking tour. Our evening ended with a special viewing of the observatory that was built for him outside the walls; surprisingly, Gauss was a professor of astronomy, not mathematics. I found poignant the story of his daughter Joanna, who devoted her life to the care of the four younger children and her father after the death of Gauss’ two wives. Joanna lived with and cared for him until his death, never marrying. This story reminded me of the devotion of astronomer William Herschel’s sister Caroline, who similarly dedicated herself to her brother and his craft, to her personal cost, although she was ultimately recognized as a renowned astronomer in her own right.
This August was the 25th year of the Methods in Computational Neuroscience course at the Marine Biological Laboratory in Woods Hole, MA. This intense four-week course has fostered many careers in the field. The course was founded by Jim Bower and Christof Koch, and has changed directorship every five years: David Tank and David Kleinfeld, Bill Bialek and Rob de Ruyter, Bard Ermentrout and John White, me and Michael Berry. The directorship will be taken over in 2013 by Mark Goldman (pictured above in the illustrious Class of ’97) and Michale Fee.
This year’s class had several special features. Lectures and celebrations brought almost all of the past directors to visit; Jim Bower memorably embellished many of the grand stories of the early years of the course. Greg Gage of Backyard Brains ran a practical session in the first week where students constructed a miniature neurophysiology rig. The Backyard Brains outfit stayed on throughout the course working on upgrades to their home physiology kits and shared the results with the students and course kids: new electrodes to measure EMGs and suction electrodes to record from jellyfish. We were amazed and delighted to see calcium spikes from Venus flytraps and to watch squid chromatophores dance to iPhone signals. The correlated variations in chromatophore size recorded in Roger Hanlon’s lab were the subject of Emily Mackevicius’ course project. We concluded the course with four talks aimed at pointing toward the future of computational neuroscience from Sebastian Seung, Bill Bialek, Larry Abbott and Haim Sompolinsky.
25 years truly is a milestone for the field. Many departments and neuroscience programs now include theorists, and a growing number of experimentalists are adept both in physiology and in analytical approaches. Despite, or because, of this, the rapid experimental advances in in vivo recording, imaging and optogenetic manipulation, high-throughput physiology and anatomy are posing deep new challenges to bring theory closer to experimental reality, while still seeking fundamental underlying principles. As said Haim: “We are approaching the end of the golden era of ignorance.”