Top-Down Learning — An Approach to Understanding Machine Learning
If you’re like me and like building things from scratch but as soon as you start getting into all the details, fear of imposter syndrome comes rushing in. However, a top-down approach towards learning any new complex subject matter can help you get through it. Top-Down is a very general term that can be applied to a wide variety of fields so let’s start with a formal definition. According to the Merriam-Webster dictionary, Top-Down refers to the breaking down of large general aspects (as of a problem) into smaller more detailed constituents : working from the general to the specific.
In the world of computer science, a top-down approach plays a key-role in software development. Before one sits down and begins coding, it is essential that there is a sufficient level of understanding of the system that you’re trying to create. For example, let’s say you want to become a professional musician but don’t know how to play an instrument. A top-down approach would be to first learn enough chords on the guitar to play your favorite songs. Then once you become familiar with the way your fingers move across the fret board and you’re comfortable with the feel of the instrument, you decide to dive deeper into music theory to understand how each note interactions with other notes. The same approach can be applied to understanding machine learning. Understanding how each regression and classification model works under the hood requires a deep understanding of linear algebra, statistics, and calculus. Fortunately, you don’t need to have a PhD in mathematics to run a random forest model. So if you are thinking of enrolling in a data science bootcamp but are feeling intimidated by the math portion, it’s important to manage you’re expectations on what knowledge you’ll graduate with. 15 weeks is hardly sufficient time to master any skill, especially in a field that’s as large as data science.
Research
Recent research in cognitive neuroscience has shown how a top-down approach enhances the learning process by stimulating areas of the brain that contains lexical knowledge. So when we learn a new word, top-down processing allows us to connect that word to concepts that we already know. In a study done by Deborah Lovich in 2004(https://www.nsta.org/publications/news/story.aspx?id=53144), researchers took a sample of 48 students and split them up into 6 groups. Each group was given a set a questions and asked to create a presentation on how to approach the questions. Using a neighboring lab group, students tested their plan on a second set of questions and asked to evaluate it on a scale of 1 through 5 (inadequate to exceptional). The next next day, the same group of students went through a metacognitive lesson before being given a third set of questions. Each group was then asked to apply the concepts they learned in the lesson to the next set of questions and go through the same process with the neighboring lab group as the day before. The results concluded that the score in the inventiveness strategy category had increased, which illustrated how important the role of metacognitive awareness plays in the classroom.
Real-Life Application of Top Down Learning
Big Picture Learning is one example of an organization that takes advantage of this style of learning. Created by Dennis Littky and Elliot Washor, this organization’s mission is to use project-based learning to help identify a student’s interests by applying them to larger projects. This is a similar strategy of intensive Data Science Bootcamps such as Flatiron school where the goal is to teach data science and machine learning through individual projects that are applicable to real-world problems. Once you become familiar with all the new terms and concepts, it becomes easier and less daunting to dive into the gritty details of how a machine learning model works underneath the hood and how a slight adjustment in a hyperparameter may affect the final results.
