Meet the CoSMo organizers
Sensory and Movement science are quickly absorbing techniques from various computational disciplines, including Bayesian statistics, Systems Identification & Control, Machine learning, and Causal Inference. And, indeed, in all these disciplines there are text books, online resources, and summer schools that teach the techniques. However, the concepts, how these tools relate to reality, how the theories relate to data, what the relationship between those disciplines is, and how the experiments speak to real questions are generally taught as afterthoughts, if at all. This is where CoSMo is unique. Across all the popular techniques we primarily focus on the understanding they can produce, and on the many different kinds of understanding that can be enabled by good computational work. CoSMo trains scientists to seamlessly blend disciplines, have clarity about their large scale objectives, and lead the sensory-motor community in new, promising directions.
Participants will be strictly limited to 40 to ensure an excellent instructor to participant ratio. As the organizers, we are confronted with the difficult task of selecting the participants from all qualified applications.
The organizers will evaluate all applications based on the submitted material. We are particularly interested in understanding your motivation for attending this summer school; how it would impact your career and how external referees judge your potential to meet those career goals. In order to ensure that everyone benefits as much as possible from this summer school, we will also take the appropriateness of your educational background into account. The selection of participants will therefore be based upon how much they would benefit from this training opportunity.
CoSMo 2020 will take place at the University of Western Ontario in London, Ontario, Canada.
Participant housing will be on campus. More details soon.
London, ON is accessible by air, train, bus or car. More details soon.
Cost will include, accommodation and 3 meals/day (but not travel!).
When applying, you will need to upload your CV, statement of interest and 2-3 reference letters!
You also need a Google Scholar account!
We are very pleased to announce the preliminary program
below. Lectures will be organized in themed modules with
two or three interacting lecturers. Morning (9am-12:30pm) and
afternoon (1:30-5pm) sessions will be a mixture of
tutorial-style lectures and hands-on Python simulations using
Google Colab. In addition, there will be professional
development lectures and an introduction to data / model
sharing. There will also be a 2-week long project where
attendants in small groups can readily apply the newly acquired
computational tools to a research project (e.g. re-analyze data
from the data base, build your own models, etc...).
All lecture materials will be available here...
|Aug 3-8, Aug 10
(1) Fitting models to understand Neurons
(2) Linear Systems Theory (saccades) & Kalman
(3) Fitting models to understand Populations and Behavior
(4) The Bayesian Brain
(5) Causality and Interpretability
(6) Motor control
(7) Reinforcement learning
|Aug 4 (evening)
||Konrad Körding||Paper writing 101
|Aug 5 (evening)
|Aug 6 (evening)
||Konrad Körding||Grant writing 101
|Aug 7 (evening)
|Data / model sharing
Aside from lectures and hands-on tutorials, there will also be social and professional events. Those will permit the participants to do professional networking, discuss potential collaborations, exchange experiences and - most importantly - have informal contact with the lecturers and organizers. These events will include an opening (evening of Aug 2) and closing reception (evening of Aug 15), a weekend outing (Aug 9), as well as daily group breakfasts, lunches and dinners. There will also be organized on-on-one meetings of participants with lecturers of their choice. Details about these activities will be sent to all participants prior to the start of the summer school.
In the past, we have mainly focused on traditional modeling techniques and approaches that are popular within the motor control field at large. We now propose to steer the field by showcasing how cutting-edge advances in machine learning and causality research can be integrated with state-of-the-art movement models. This will be integrated with our past teaching and how-to-model innovations to emphasize interpretability of computational work that is based on modern data. Specifically, we propose to provide tools for the community that will facilitate transitioning to data science and deep learning via teaching CoSMo in a hands-on Python tutorial-based format. Established mentoring approaches of the organizers during 2-week small group projects creates a unique integrated research network what already contains over 300 trainees and 20+ lecturers. We also extensively focus on policies, procedures, processes: we teach the meta-science approach that translates to all science! We transmit this unique skill set and how to creatively use it to evolve in today’s climate of accelerating technology and discovery. We teach core research skill development that spans disciplines, processes and procedures.
We are improving the CoSMo experience along several essential axes to make it not just a great training experience but to help CoSMo move the field into the right direction. So we are introducing a range of changes. First, we switch CoSMo from Matlab to Python. It is the right time as many labs are slowly switching. It is also important as the post-academic job market is far better for Python coders than Matlab coders and CoSMo trains a number of data scientists. Second, we move data science and deep learning to the center, on equal footing with traditional modeling techniques. We still teach the same scientific insight-based curriculum, it’s just that now we use machine learning as a tool to advance science where appropriate. Third, we produce clarity about the exact roles that modern machine learning fills in the field and its pitfalls and problems. All these changes are carefully measured; we have written papers on the underlying ideas.
Every day, CoSMo teaches its participants at three levels. It does skill training, where students learn the state-of-the-art of computational approaches. It does interpretational training, where students learn which techniques allow them to ask which questions. Lastly, it teaches core competencies - how to find and abandon topics, how to write papers, and how to approach a career. Trainees also learn how to understand the various disciplines that make up the CoSMo community and their relationship.
This summer school aims at propelling promising students into world-class researchers.
See CoSMo WIKI for past Matlab-based teaching materials.