Building a neural network using just NumPy: Summer 2020
As I discovered my passion for Machine Learning (and more specifically, Neural Networks) I learned that the way
most people approach the subject is to use Tensorflow or Keras to create a powerful network to predict something.
While I'm interested in the performance and simplicity of using tools like this, I wanted to get a deeper
understanding of how a neural network
really works. I found
this awesome
book that teaches you how to build neural nets in Python using just NumPy. I spent time diligently reading this
book, discussing implementations with peers, and developing my code. At any rate, I implemented and fully
understand just about every single line of code. From optimizers, to backpropogation, its all in NumPy! Feel free
to reach out if you have any questions. Also I used this to develop a model to predict significant coronary disease
in patients, using
a cardiac
catheterization dataset.
I achieved a prediction accuracy of ~75% for a binary classification task.
(
Github link for this project).
ProjectX: March-November 2020
I worked with a team at UofT AI to design, schedule, and prepare a 3 month long machine learning research
competition aimed at incentivizing undergraduates to get involved in ML research. I worked on this project essentially
from inception (in February-March of 2020) to execution (Sept-Nov of 2020). While my role was technically
'External Relations Executive', I wore many hats. Early in the process I worked with 2 other students to reach
out to dozens of universities and colleges across North America to field teams. This was especially difficult
as the Covid pandemic was just getting into full swing, so CS departments at these institutions were preoccupied
with going online. However, we were eventually able to register over 20 teams (including teams from MIT, Stanford,
Carnegie Mellon, and UC Berkley to name a few). I also worked to find folks in academia and industry who would
donate their time to guide the teams, and as a result I secured nearly all of the (UofT AI provided) mentors for
this competition. At the end of the summer, I was given a new role: 'Director of Logistics'. In this role I helped
lead the charge in interfacing with teams (particularly answering their questions throughout the competition) and
ensuring submissions were submitted, anonymized, and assessed by the judges. You can see our website
here.
We were also featured on
The Strand,
The 'Everything Under the Sun' Podcast,
as well as by the
University of Toronto itself.