On this page
A Sneak Peak Into My Life 🔍
Welcome to my November/December personal newsletter! I'm looking forward to sharing all the amazing things that have happened over the past few months. If you're new to this newsletter, allow me to introduce myself: I'm Dev, and I have a deep love for AI and its potential to revolutionize various fields. Recently, I've been exploring the intersection of AI and healthcare, specifically in the realm of medical imaging.November and December have been filled with amazing experiences, from finishing my first semester at University, to judging at my first Hackathon to focusing on researching and building in the AI space. These last couple months have been incredible and so, without further ado, let's dive into all the exciting things that have happened over the past few months!
Judging at my First Hackathon! 👨🏽💻
Over the course of the last couple years, I’ve had the opportunity to participate in many hackathons, but this time around, I was given the chance to judge at my first hackathon! I want to extend my gratitude to the TKS directors, Niloo and Skylor, for giving me this opportunity to judge at the hackathon. It was a full circle moment because I remember being in the shoes of the participants and being anxious and nervous leading up to the presentation time. One of the most rewarding aspects of the experience was seeing how much the participants had learned and grown through the hackathon process. The projects were quite diverse and it was refreshing to see the limitless applications that AI still holds in the real world. Overall, judging at my first hackathon was a truly enriching experience and I’m excited to get back into participating in hackathons once again in the new year!
Delving back into the AI space 🖥️
Over the course of these last 2 months, I’ve been working on getting back into the AI space by reading papers about topics that I’m interested in. For some context, I’ve spent the last year and a half immersing myself in the AI/ML space, reading, writing & building cool projects with the many unique subsets of Artificial Intelligence. Recently, the AI and Machine Learning space have been booming with a ton of new research emerging. I was super curious about the findings that have come out to the public and I wrote a couple articles on medium outlining the research that has been done. The research has been promising and it definitely have the potential to drive significant innovative change. Let’s jump right into the articles 👇
Expressing Neural Networks as Decision Trees! 🌳
The first article that I wrote was definitely the most challenging one to write. In all honesty, the paper that I based my article off had a bunch of math that was hard to grasp my head around at first and I found myself just staring at the paper with the most confused look on my face 😅. However, eventually, once I got it, everything started to click and I couldn’t stop myself from delving deeper! As I write more and more of these articles, I’ve started to learn to appreciate the entire learning process even more. Before I outline my article, for those of who are interested, here’s the paper that I based my article on:
Neural Nets as Decision Trees Paper
Now let’s jump right into the article! For some background context, neural networks have a “black box” nature and thus we aren’t aware of how every single neuron works together to arrive at the final output. However, to get around this issue, it’s possible to take any neural network with any activations and express it as an equivalent decision tree.
As a result, the tree representation doesn’t limit or require modifying the architecture of the neural network in any way.
In the article, I outlined how each component of the neural network can be expressed as a decision tree equivalent (normalization layers, activation functions, etc.). The most promising part of the paper is where they tested the decision tree equivalent on a couple of datasets and the results were extremely promising.
They first tested it on a regular parabolic function. The results were great as the decision tree and the original Neural Network had very similar results. For visualization purposes, here’s what they looked like:
There was another example with half-moon classification as well and not only was the decision tree very similar to the neural network, but the decision trees also have a computational advantage over neural networks 🤯. If you want to read more, check out my article below 👇
Back-Propagation No More? Breaking Down the Forward-Forward Algorithm
The second article I wrote was a little easier to write compared to the first one, but the entire process was a huge learning experience for me as well. To give some context about this article, a couple months ago, a paper was released where they were able to simulate neural networks without backpropagation. Up till this point, backpropagation was a vital component of neural networks; the formal definition is as follows:
Backpropagation is an efficient algorithm for learning the weights and biases of a neural network.
However, this paper found a way to get rid of the backpropagation aspect and use 2 forward passes instead, better known as the Forward-Forward algorithm. This was really a ground-breaking discovery because backpropagation has its flaws, but we’ve always used a more inefficient and expensive way to get around it. If you want to read the specifics of the paper, click the button below 👇
Forward-Forward Algorithm Paper
Moving onto the article that I wrote; in this article, I broke down how Neural Networks work and showed where the limitations exist in modern-day neural networks. I also highlighted why current methods aren’t as effective and why the Forward-Forward algorithm holds the answers to the future of Neural Networks.
The Forward-Forward algorithm was tested on a bunch of different datasets, the most popular ones were MNIST and CIFAR-10! After running the experiments on the dataset, it was clear that although the Forward-Forward algorithm might not be as efficient as backpropagation, it still has the ability to deliver similar results, just with more epochs! In the article, I showed a variety of different applications of the Forward-Forward algorithm, ranging from regular pixelated/static images to video inputs! Check out my article below 👇
Forward-Forward Algorithm Article
MNIST Project
Alongside writing articles about the new discoveries and findings in the AI space, I’ve also jumped back into making AI projects as well. I spent parts of December going over the basics of Tensorflow and Convolutional Neural Networks to grasp a better understanding of the foundations of AI. This was more of a ‘warm-up’ project, but I used Tensorflow and Numpy to create a Convolutional Neural Network that would be able to classify hand-written digits. This was just the beginning, I’m looking forward to diving further into the Computer Vision space over the course of this new year! Be on the lookout for new projects :) If you want to check out my code, I’ve linked my github below 👇
Expressing Gratitude 🥳
It’s finally 2023 and I wanted to end off my final newsletter of 2022 with a small reflection of the entire year. As I look back on the year that has just passed, I always take some time to reflect on what I’ve accomplished, the challenges I faced, and the things I might want to do differently in the future. It gives me the right momentum going into the new year and allows me to start the new year off on the right foot, with a clear sense of purpose and direction.
Personally, 2022 was full of unexpected twists and turns and I found myself dealing with a lot of change. I went from a person who was in his own world, working on projects by himself to someone who got the chance to work on the world’s most challenging problems in the FinTech space with Interac. In between all that, I had the opportunity to work alongside IKEA and pitch our solution the board of executives, build a moonshot company, start my freshmen year at UofT, and build genuine connections with people!
The year was full of many humbling and not-so humbling experiences. One of the crucial takeaways that I walked away with is that the process is always more important than the product.
Process > Product
Often times, it’s difficult to push out the final product and there’s a breaking point where sometimes we just want to get it done and over with. But throughout this year, I’ve taken part in many projects, some short, some long, but as I’ve taken part in these projects, I’ve developed a true passion for the entire process. At the end of the day, the product isn’t what is going to help me grow, but it’s the entire process that will! Many experiences throughout the year really helped internalize this lesson, whether that be during my internship at Interac or building an AI project! Now, going into the new year, I’m going to be intentional about doing projects for the process and not just for the outcome/product that will come out of it.
Before ending this section off, I want to extend my gratitude to my family, friends, colleagues, directors, mentors, managers and everyone else that have stuck alongside me throughout this year. It was one of the most challenging, yet rewarding years of my life and I’m grateful to have had all of these amazing people be a part of that journey!
What lies ahead… 😤
That marks the end of my bi-monthly update for the months of November and December. Now that we’re at the end of my update, in typical newsletter fashion, I’m going to sign off by highlighting some things to look out for during my next newsletter 😁
Continuing my Research in Data Science x Medical Imaging under Prof. Tyrell @ UofT 🧑🏽🔬
Building more projects in the Computer Vision space!
Writing in-depth & technical AI articles about recent papers and advancements ✍🏼
That’s all from me! I hope everyone’s last couple months went well. Thank you for spending time to read my newsletter and I’ll see everyone in the next edition!
If you enjoyed this, feel free to share and subscribe to my newsletter by putting your email down below 👇. Thanks for reading, I’ll see you soon!