Neuromatch Academy Experience
Would you think that an online and virtual summer school could be as effective and efficient as a real on-site one? We did not think that would be the case when we applied for the interactive track of Neuromatch Academy Computational Neuroscience Summer School. However the case was the absolute opposite of what we thought. The experience Neuromatch Academy provided us was a really joyful and efficient one in which we learned so much while we were able to point out our questions that are answered quickly.
We are Bora Çelebi and Şule Taşlıyurt from Yeditepe University Cognitive Sciences Master’s Program. We completed our lessons and we are actually at our thesis year. We took part in the Neuromatch Academy Summer School which was held between 13–31 July 2020.
Thanks to our professor Funda Yıldırım, we were able to get the announcement about the registration about Neuromatch. The Academy provided students with two different choices; observer track and interactive track. The observers were able to access Q&A sessions, Youtube lectures and IPython notebooks and in additive materials that were available generally in weekends. However, they pointed out that the interactive ones would be enrolled in a group in which they would be guided by teaching assistants in lectures and mentors for their group projects. While this structure for interactive students seemed perfect as an onsite summer school, we were thinking that this would be a bit better version of a MOOC as these type of organizations were new and not so many institutes were able to get professional in this mere time window where all of the things went online due to COVID-19 pandemic. But the reality was different and we really appreciated the hard work and the time the organizers and volunteers spent in making this a real summer school.
Firstly, the interactive track was constructed in a way that some number of participants create a pod accompanied by one teaching assistant. Our pod was consisted of five people. Our joyful T.A. [Başak Kocaoğlu] had been appointed to us 2 weeks prior to the start of the program. She contacted us immediately and organized a mailing list where we introduced ourselves and got to know the other fellows in our pod. As far as I know, ours was the only fully Turkish nationality pod even though the instructing and communication language was English. After this assignment to our pods, we were introduced with the Neurostars platform which is basically a forum where people can discuss and exchange experiences and questions. Then there starts the exhaustive 3 weeks marathon of Neuromatch Academy.
The timing for the summer school was like we were supposed meet at 14.30 to discuss our project -will cover that soon- and then at 18.30 the first Q&A session starts. After some breaks, at 21.00 the lecture session starts that would generally last four to five hours according to the intensity of the content of the day. So, in these dates, whole lot of the time of the day was spent on NMA for us which made it a perfect environment and structure to develop our skills and learn new content.
In the first few days there was not to do with the project at hand so until Wednesday we went with the lectures generally. The lectures program mainly consisted of two things: videos and IPython notebooks. For everyday there was an introduction video which covers the basic knowledge and the ground idea of the subject of the day with few examples of real-life research that used the technique regarding the subject. Also, some fundamental mathematical and theoretical structures were given by the introduction video which generally lasted about 30 minutes. By the way, all videos and notebooks are available on the internet that can be accessible for everyone. And then at 21.00, all the pod members meet in a pod specific Zoom session and start to go over the notebooks together with peer programming and discussion at focus. So, the general structure was something like to go over markdown cells which cover crucial information on the subject, watching relatively shorter videos about the portion of the content, looking at the code cells that were comprised of fundamental functions and code brackets that tries to show how to do it and then exercises that were left for students which gave us the possibility to do hands-on experience with the subject at hand. In this aspect, peer programming was the choice of organizers as everyone tucks into the exercise trying to figure out how to do it and then coming into a convention that works. Someone always explains how it is done so everyone could be at the same page about the practical implementation of the content on Python. An important side note here; as there are vast numbers of possible packages in Python that would facilitate the coding from tens of lines of code to just two or three, NMA was not about that. The notebooks provided only used numpy and scipy (and pandas to facilitate organizing the data) that are the bare minimums to make mathematical computations fast and efficiently in Python. So, the curriculum showed all of the computations and formula from ground up and made us explore it in low-level. That meant a lot in the way of learning concepts of computational neuroscience and how to apply it as exercises made you write the code from formula without any specific packages. Therefore, It was fascinating that the structure of these lectures were well planned for a virtual summer school, and appointed T.A.s (at least ours) were very prominent in her/his field and helped us a lot while going through the notebooks and lectures.
The curriculum of NMA was well established. It started from the main concepts as different types of modelling approaches in order to introduce how neural systems can be modelled in different ways according to the data and the neural functions to a phase that every important modelling approach was introduced with coding samples that would give you an important glimpse on what is happening in academia. The program started with the modelling types with explaining “what”, “how” and “why” models cover and then evolved into showing every specific type of models with their own practices in neuroscientific research. The curriculum can be seen in the picture below. From linear models to dynamic networks it was really exhaustive trying to cover all of the main subjects in computational neuroscience. On the other hand, most of the data provided in the notebooks were single neuron or microelectrode array recordings instead of the signals gathered by the neuroimaging data. So that can be a good or bad according to the point of view. In this aspect I can say that the only complaint I had about the NMA was the lack of examples and lectures about the implementation of certain models on neuroimaging data.
The other important part of the interactive track of NMA was the group project. Generally participants in the pods are divided into groups of three to four people for group project because of our pod was 5 people we stuck together for the project. First we were supplied with six datasets from every part of the neuroscience field including both animal and human studies. There were cell recordings and also fMRI datasets available. So first week we came up with a research question on HCP (Human Connectome Project) dataset that is consisted of fMRI data on some particular tasks. According to the research question and the choice of the dataset, a mentor was assigned to the project group who would help us refine the question and find ways to model the data we were given. Every mentor was really proficient and known figures in the field but we were lucky to be supervised by Professor Aina Puce. She made us create a great plan from research to hands on modeling so that we made through the project in a planned way. Dr. Puce was very helpful in this aspect and wanted us to feel the full experience of doing computational neuroscience so she showed us the steps into a complete project. We looked at the interaction between mPFC and anterior insular cortex using Granger Causality. In the way of finishing the project we needed to clean the data, normalize it in some ways, did the GLM and then did the causal inference. So the project part of NMA gave us a good chance to explore, process, analyze and model the data at hand according to the techniques available in the course program.
All in all, contrary to our prior belief about how this experience would be, we passed a great time learning so many things either in lectures either in hands on exercises and project. It was a really tiresome track which made us spend lots of the day studying according to the program but that is one of the things that made NMA like an onsite summer school. We would like to thank all of the organizers, mentors, TAs and volunteers that made the academy a great experience.
Link to Github repository for all of the content: https://github.com/NeuromatchAcademy/course-content