Neuromatch Academy: An experience report
Table of Contents
This summer, I participated in Neuromatch Academy (NMA) Summer Online school as a TA for Computational Neuroscience (CN) and a student for Deep Learning (DL). It's been several months away since I submitted our group's final DL project for NMA, but what happened those 7 weeks are still vivid today.
Taking the NMA Online course was like eating a box of chocolates with a plethora of flavors. You never know how different tomorrow's chocolate is going to taste. You must be prepared to learn knowledge from all kinds of areas. For CN course, it covers materials not only in neuroscience but also from information theory, machine learning, signal processing. For DL, knowledge from computational version, natural language processing, reinforcement learning, and so on were digested by us in 3 weeks. You could choose to participate in either CN, DL, or both.
NMA encouraged everybody to finish a project by the end of each course. This might sound intimidating. In fact, I heard that many groups failed to submit a proposal for the final project before the deadline. So, the deadline was extended. Still, many of them ended up completing amazing final projects. Keep in mind that, we do not need to master any of the fields in order to finish the project, NMA also provides nice templates for us to use. Additionally, many project ideas come from the previous project which is organized well on NMA website. Only experts could link knowledge from all kinds of areas together, and it is a lifelong process. For this course we just demand to understand what we need so far. ### Suggestions for Beginners For beginners, it is both rewarding and challenging. You have to be self-motivated enough to finish the course. If you have only a little background in math and computer science, NMA provided us with some precourse material to learn before the summer. I highly recommend any beginner to watch the books, the video and do some practice during your free time. Those materials are fundamental for you to understand the course material. And if you fail to understand some key materials, you are likely to lose your way in the following courses.
I hope this did not scare you away from participating in NMA. That's all the hard part about NMA. Once you handled it, you are going to have some wonderful swim in the sea of knowledge. You will be split up into different pods paired with one TA. The daily course consists of several short videos followed by some programming exercises. You have 3-4 hours to learn the material within your pod and the next 3-4 hours to do the final project with your group members. NMA provides many interactive demos for you to play in the jupyter notebook. For example, you can visualize what principal features learned by your neural network. Many tutorials included widgets for you to adjust the parameters in one function, and you will see how the output will change immediately. NMA also provides daily quizzes (around 4-5 questions) that help you to check your understanding. Don't worry, your performance won't influence whether you would receive certification or not.
1. Suggestions for TA
As a TA, I knew some of my students might not want to answer questions or even ask questions because they have no clue about what I am talking about. Sometimes, discussion with several random group members before revealing the answer together as a whole group is beneficial. Therefore, I coded up a random group generator to help them. You might say, zoom has this function already. But most of the time, I also use it as a way to generate a schedule of who will answer which question. Just by setting how many groups you want to generate equal to the total number of people, you will have a number before each student's name. (you can find the code on my GitHub). I showed this schedule to my students before class, so they could prepare the answer well. If they fail to answer the question, I will provide more sub-questions to guide them. It is helpful when we articulate our thinking and sometimes the answer will become oblivious as we are speaking.
When you use the random group generator, make sure to put your students' names into the variable studentname.
var student_name = "Apple Zeng\n"+ "Mango Wu\n"+ "Pear Zhang\n"+ "Avacado Gao\n"+ "Potato Han\n"+ "Banana Chen\n"+ "Tomato Li\n"
I always remind myself and my students that "You are not alone"! We can also discuss questions with TA or classmates in the zoom breakout room. Plus, there is a discord channel for us to connect with students from all over the world. It was strange for me to have a discussion with students from Iran, Turkey, and Greece about how to be a good TA. Even though some of the countries may have different political ideologies, in that room, we are just students who love computational neuroscience. You can always post questions in the discord, the faculty members who designed those tutorials will sometimes answer your questions, or they may realize there are some mistakes in the material, HAHA.
2. Suggestions for TA
In the end, I want to say, completing NMA might not provide me the ability to finish one project from scratch. However, it gives me the courage to read papers full of formulas that I don't understand at all before, and it teach me some skills to search for information that will tackle hard questions bit by bit.