David used Traverse to master machine learning fundamentals and stay on top of the latest research
- Traverse helped him get fluent in the mathematical foundations
- Traverse provided him with a system that gave him the confidence that he’ll remember
- It also has made it more fun to deeply learn difficult topics, resulting in 3 publications and an invited talk to the Meta Reality Labs Research Audio Team
David is a research masters student in machine learning at Carnegie Mellon University.
With AI models like ChatGPT being all the rage, machine learning is a hot topic. But machine learning is also a very challenging topic to learn. It combines various fields, most notably a math, statistics and computer science.
David noticed that he didn't retain some of the mathematical fundamentals as well as he would like. Some fundamentals had gotten rusty as he hadn’t taken those courses in a long time. And in his research master’s, classes often jumped around from one topic to another. David would love to go more in depth, but exposure in class never seemed quite sufficient to truly synthesize a topic.
Before starting his master’s, David had never consciously used structured learning methods. Since he didn’t really have any system for studying, it felt haphazard as to whether he’d remember or not, and he was uncertain as to how to approach learning new things.
David first heard about flashcards from a friend. He’d come across Notion and Anki but he didn’t really like neither. When he came across Traverse, the interface just made sense, and the promise of learning how to learn resonated, so he gave it a try.
He used Traverse for two main purposes:
- Revisiting the fundamentals of probability, linear algebra and the other main fields of machine learning
- Remembering research papers, literature and other things he reads and listens to (like the Andrew Huberman podcast)
Below is his map of learning probability and linear algebra. He went through the Kevin Murphy Machine Learning book, as well as all probability problems listed on Wikipedia (like the famous Monty Hall problem), and grouped them according to their mathematical properties.
Then, for each of the problems he added flashcards to remember the problems properties and deduct the way to solve it.
He did the same for programming fundamentals, this time going through the problems of LeetCode and learning the right way of solving each.
Besides the fundamentals, David also reads many papers to stay on top of the latest research. If he wants to remember a paper, he’ll add them to Traverse. An example is the research on Natural Language Processing (NLP), which is the field that underlies models like ChatGPT. He’s divided the field into its main topics, and added flashcards to remember the main research insights for each.
Traverse provided him with the confidence that he’ll remember all of the fundamentals. Now he has a system he can rely on. When he adds stuff to Traverse, makes connections, and reviews his Traverse, he know he is going to remember. He also noticed that he is getting more fluent in math after he’s reviewed it in Traverse.
David used Traverse as part of a line of research resulting in 3 publications and an invited talk to the Meta Reality Labs Research Audio Team.
He also used Traverse to prepare well-received internal presentations on Instruction Tuning and Mechanistic Interpretability to the MLSP (machine learning for signal processing) Lab at CMU.
And finally, David found that the additional confidence of remembering makes it more fun to learn difficult stuff, an experience similar to what “tools-for-thought” and fellow AI researcher Michael Nielsen describes.
Traverse is a learning tool which helps students and researchers learn better using scientific learning techniques. Our values are the search for truth, simplicity, joy and connection. We’ve helped over 10.000 people learn better so far, learning languages, medicine, computer science and other subjects.