For graduate students, postdocs, and young faculty, attending a summer program is a great opportunity learn something new and make new friends. But there are many summer programs that neuroscience researchers can choose from and deciding which one to attend can be difficult.

Here I outline my experience with two of the summer schools that I had the privilege of attending: Neurohackademy at University of Washington in Seattle and Methods in Neuroscience at Dartmouth (MIND).

TL;DR; Check out Neurohackademy if you want to focus on learning about good programming practices, wide range of analytic tools, and an introduction to hacking. I would recommend MIND if you want to focus more on learning advanced analytic methods that you were curious about or wanted to apply to your own data. The topics offered at MIND change from year to year but a core theme seems to be methods useful for analyzing naturalistic and multidimensional data with temporal, linguistic, and interpersonal dynamics. Personally, I would recommend going to Neurohackademy in the first/second year of grad school and then to MIND in later years.



Neurohackademy

Neurohackademy is a two-week program focused on neuroimaging and data science hosted by the University of Washington eScience Institute in Seattle. Participants come from a wide range of backgrounds including graduate students (master, phd), post-docs, as well as those who work in the industry. A detailed course schedule can be found here but in summary the first week is devoted to introductory tutorials & lectures and the second week is reserved for hackathon projects.

Lectures in the first week covered introductions to common programming languages including Python, R, and Matlab. Tutorials for each language covered both basic concepts and use in advanced analyses. For example, tutorials in R included data munging, visualizations, as well as using Neuropointillist for fMRI analysis in R. Topics in Python were the most extensive including data munging, visualizations, packaging, software testing, fMRI analysis using nipype, and deep learning using Keras. There is also a big focus on open and reproducible science, and thus covers using open science tools such as Docker containers.

My favorite tutorials were deep learning with Keras by Ariel Rokem and introduction to web applications by Anisha Keshavan. Ariel did an amazing job explaining the fundamental building blocks of a neural network and how they come together to form more complicated models such as DCNNs with an easy to follow tutorial. Anisha’s web application tutorial described the fundamental building blocks in building a web app while also touching on popular tools such as Vue.js and Firebase.

The hackathon hosted more than a dozen projects with diverse topics using web applications, cloud computing, open datasets, and deep learning. I participated in two projects: hacking the Study Forrest data and building PaperWiki. Using the Study Forrest dataset, we looked at how brain representations of characters evolve over time across individuals. We tested whether participants generated different representations of characters over time due to personal preferences or identical representations over time due to shared information. On a whole brain level, we found evidence for for the latter for the main characters Forrest and Jenny but not for the peripheral characters. Check out the presentation video below:

PaperWiki was designed to be like Wikipedia for research articles. Whenever I read academic papers, I wanted to know and discuss what other people thought about the paper, the history of the paper (e.g., why is this paper meaningful?), and any updates regarding the contents of the paper (e.g., was there any issues in the paper?). This kind of information is often scattered and surprisingly difficult to get and that is why I think PaperWiki could be a useful tool for researchers as well as students. Currently, one can search for articles and create a Wikipedia-like article for each paper in markdown. There is currently a Disqus commenting board which I hope to replace with a reddit-like board in the future.

Overall, Neurohackademy is a great opportunity where you can learn a ton about programming, neuroimaging research, and the open science community by meeting and interacting with core developers of the Python scientific computing ecosystem such as numpy, sklearn, nipype, and Jupyter. If you are interested in what the lectures are like, checkout the lecture videos here.

Lastly, summer in Seattle is beautiful and here is a picture of the attendees on a cruise to prove it.



Method In Neuroscience at Dartmouth (MIND)

The MIND summer school lasts about a week and a half but the shorter duration still packs a tons great lectures and tutorials. The hackathon at MIND begins in the first few days, and the days are structured to have lectures in the morning, tutorials in the afternoon, and working on hackathon projects in the evening. I would say that the MIND schedule is a bit more intense than Neurohackademy as you end up working on projects in the evening whereas evenings in Neurohackademy didn’t really get fired up until the second week of hackathons.

There is a day devoted to learning about open science tools such as Git/Github, containers (Docker/Singularity), and Jupyter Notebooks but the time spent on these topics is relatively less than what you would get from Neurohackademy. An interesting comparison is that MIND uses the MIND-Docker container for tutorials whereas Neurohackademy uses the Jupyter hub. I think both options are great for creating an easy-to-use and uniform environment for tutorials but the hub approach seems a little more accessible because you merely need to enter the website address without having to install a ~10gb docker container.

The best part of MIND were the lectures and specific tutorials aimed at data analysis. When I attended in 2017, we learned about spike decoding, fMRI decoding, network dynamics, spatiotemporal models, HMMs, and much more. This year included additional topics such as RSA, natural language processing, and conversation analyses which I was sad to miss. As the focus is on learning new analytic tools and methods, the hackathons projects also lean more towards applying these methods to your own or public data (in contrast to Neurohackademy where more people focused on building and improving tools for analysis and visualizations).

When I attended, I participated in applying a negative affect classifier, PINES, to neural data from Aaron Heller’s study on looking at corrugator activity while viewing negative/positive images. We found that PINES predictions correlated well with arousal and negatively correlated with valence although the predictions weren’t so good when we tried to corrguator activity or win/loss trials. Also, after training a whole-brain ridge prediction model for valence and arousal we compared the cross prediction accuracies which turned out to be fairly high suggesting that these two modalities may be similarly represented in the brain.

Overall, I think MIND is a great summer program especially for individuals who have some experience in programming that want to focus on learning and applying new methods in a short amount of time. You get to interact with instructors to learn those new methods from the experts who use them. You can check out the MIND lecture videos here.

Also summer in Dartmouth is not too bad either!

The verdict

If you are early in your graduate program or if you would like to get a birds-eye view of the programming landscape used in neuroimaging analyses with a focus on open data science, I highly recommend checking out Neurohackademy first. If you are a more seasoned programmer or have specific topics you would like to learn about such as spatiotemporal analyses, natural language processing, RSA, neural decoding, and network analyses, I would highly recommend checking out MIND.