An intensive AI/Machine Learning Bootcamp for high school students
MehtA+’s Machine Learning Bootcamp is a 6-week virtual summer research bootcamp taught by MIT and Stanford engineers for rising high school freshmen to college freshmen.
Edit: Please note that this article references MehtA+ Digital Humanities Machine Learning Bootcamp (Summer 2021). While some of the information in this article may be outdated, our approach to teaching still holds. Ferro Rocher anyone? ;)
In most high schools, students learn a variety of interesting material: theories posited by European philosophers such as John Locke’s tabula rasa, calculus theorems like L’Hôpital’s rule, realismo mágico in Gabriel García Márquez’s Cien años de soledad and other works of Spanish literature, Faraday’s law of induction in electromagnetism and bildungsroman theme present in To Kill a Mockingbird.
Unfortunately, as teachers are forced to rush to cover material that will appear in AP exams and satisfy state requirements, high school students often lose the connections that exist between different subjects and the reasons for learning the material in the first place. Many times, this grueling journey often leads students to feel exhausted and apathetic to what they are learning. As students go on to attend university, more often than not, they will attend classes taught by professors who are forced to focus on their research, rather than teaching. As these professors explain even the simplest concepts in a convoluted manner, students struggle and work overtime to understand the material, extinguishing their spark of learning, perhaps forever.
However, it doesn’t have to be this way! Barring any systemic changes that perhaps should take place in higher education, at MehtA+, we feel strongly that learning “university-level” concepts and completing interdisciplinary research at the high school level at a MehtA+ Machine Learning Bootcamp is an excellent way to understand why we learn what we learn in high school, to understand how different fields work together and also to get ahead!
Why AI/Machine Learning
Can we use computers to simulate human intelligence? Artificial Intelligence (AI)/Machine Learning seeks to answer that question. The intersection of biology, computer science, calculus, statistics, psychology, philosophy and linguistics, machine learning is used in many facets of our lives — from spam filters to chatbots and forecasting the weather to fraud detection, machine learning is used to automate many mundane tasks.
While we are often a beneficiary of the good that AI has to offer, AI is often made out as the villain in the media — as something that has a mind just like a human and is capable of many tasks. Fortunately or unfortunately, we are not quite there yet. While the media exhibits AI as having artificial general intelligence, so far, we only have narrow AI, where AI can only perform a single task very well. That is not to say AI still cannot be dangerous. It still can be, due to the unconscious biases of the people who have gathered training data and trained the AI model.
That is why learning AI/machine learning is a great opportunity to learn and apply a variety of subject material at once. Knowledge of the sciences and the humanities will help AI engineers understand where AI algorithms can be applied. These engineers must have a strong mathematical basis (algebra, geometry, trigonometry, statistics, calculus, linear algebra, etc.) and clear understanding of the AI algorithms — their uses and their limitations. In order to execute the algorithm, engineers must know principles of computer science and how to program and train their model. Proper data collection and understanding the implications of the resulting model necessitates knowledge of philosophy, ethics and history.
Our pedagogy
If AI indeed requires a vast knowledge of many fields, it is often hard for people to comprehend how high school students are equipped to learn a subject that university-level students learn.
Well, learning AI is not that different from eating a Ferrero Rocher — yes, you read that correctly! In case you have not eaten one, Ferrero Rocher is a chocolate ball composed of several layers — the more layers you peel, the better it gets. Just eating the first layer is also a delicious experience. Similarly, understanding the different types of models used for natural language processing, computer vision, sentiment analysis, speech recognition, etc. only at a very high level is a rewarding experience. As a person delves deeper into the world of machine learning, they will be able to further appreciate the beauty of these machine learning models.
The first layer of our Ferrero Rocher is generally a geometric representation of the machine learning model. We find that it is the easiest way for students to understand the uses, strengths and weaknesses of the model, if they are able to visualize it. The next layer of our chocolate is usually programming, where students learn how to code their model and what Python libraries they must use so they can apply the model in a real world context. Additional layers of the chocolate are the mathematical basis of the models, which may involve a thorough understanding of statistics, trigonometry, higher level calculus and linear algebra.
We hope that students gain a mastery in the first two layers. While brief explanations of relevant mathematical subjects are provided in camp, we expect mathematically advanced students to take the time to understand the mathematical proofs discussed in class. We encourage other students to try to absorb the mathematical concepts, so that when they actually learn those concepts in school in-depth, they will already know the application of those concepts.
What sets us apart (besides our very interesting analogies :)) is the time we invest in each student. We are committed to always having a 8:1 instructor to student ratio or better. Our teaching staff hold scheduled office hours every single day as well as on-demand office hours — if student needs 1-on-1 attention, we welcome them to set up an individual meeting with one of our instructors. Since our teaching staff is located all over the world, there is always someone ready to answer students’ questions at any hour of the day — or night — on Slack. We believe instant feedback is important to our students’ learning.
As the weeks go on, we find that students become more self-sufficient and use other students as resources, giving and receiving help from them. The best way to learn the material, after all, is to teach it!
Structure of the camp
Our camp meets virtually for 6 weeks. There are 4 hours of lecture per day during weekdays. Depending on the day, lectures may involve going over new concepts, reviewing homework, a talk from a guest speaker who will discuss real world applications of machine learning, talent show, group activities and friendly competitions, which test the students’ understanding. Homework is assigned almost every day to reinforce what students are learning in class.
In addition to completing the daily homework assignments, we recommend that students constantly engage with the material. For some students, that may involve perusing through machine learning papers in their spare time. For others, that may involve memeing. We have a dedicated Slack channel for AI memes, where students often post relevant memes to the material they are learning. Learning should be fun after all!
Over the course of the 6 weeks, students will also work on a final research project of their choosing, for which they must create a website, a poster presentation and a paper and present in our annual MehtA+ Machine Learning Conference.
Last year, we had 40 students from the USA, Canada, India, UAE, South Korea and China. They researched everything from imitation of the playing style of non-grand chess masters to handwriting recognition of writings by George Washington, Latin translation to music generation, early detection of Alzheimer’s to detection of “fake news”. They published their research online, in order to contribute to the open source movement.
One MIT professor who had attended the MehtA+ conference last year was amazed that “the students created such exciting and innovative projects (incorporating the latest advances in ML!)”
Digital Humanities Machine Learning Camp
We offer the machine learning camp every year, albeit in different flavors. In Summer 2021, we will be offering a Digital Humanities version of the bootcamp where students will receive the opportunity to complete a machine learning project in the humanities under the guidance of MehtA+ instructors and humanities professors from prestigious universities around the world such as University of Bologna, University of Buffalo SUNY, Northeastern, Kenyon and many more. Examples of projects that students may work on include exploring tweets for dialectic change in Spanish language and determining authorship of and dating 19th century Italian texts. Students may receive the opportunity to publish their interdisciplinary research in academic journals.
A few final words
We encourage all rising high school freshmen to college freshmen to apply to our Machine Learning camp. It is a great opportunity not only to learn a lot of advanced material, but also to meet like-minded peers from all over the world. Even after camp, our students keep in touch, often partnering up with each other to work on machine learning (and other) projects. Under our guidance, post the camp, many of our students volunteer to give lectures on machine learning to students in underserved communities as well as serve as instructors for MehtA+ computer science internships for middle school students.
While we are absolutely thrilled when our students gain a clear understanding of the various models of artificial intelligence, our greatest delight is when we see the joy of learning rekindled in our students. We hope that all our students will go on to be lifelong-learners.
