On 16 December 2022, I attended Georgia Tech’s Commencement ceremony at Bobby Dodd Stadium and walked across the stage to receive my Master of Science in Computer Science diploma.

(Since Georgia Tech doesn’t show your specialization except in your transcript, and doesn’t confer academic honors to Master students, can I quickly state for the record that I specialized in Machine Learning and had a 4.0/4.0 GPA? )

It was a proud and significant moment for me. To me, it was a closure to a period of my life, spanning 6 years, 2 master’s degrees, and the births of my 2 children.

切っ掛け (The Beginning)

I first heard of the OMSCS program back when it first launched in 2014, back when MOOCs and Udacity were still in their first wave of popularity. In that past life, I was working as a salaryman in Tokyo. I was trying to learn R on Coursera and that set me down the rabbit hole of learning programming and computer science. But my background was in accounting. So, I created a 5-year plan with the end goal of obtaining a master’s degree in Computer Science and I set out on my journey to explore this newfangled computer science thingy.

I searched around and enrolled in the University of Adelaide’s Graduate Diploma in Computer Science in January 2016, where I learned fundamental CS topics such as OOP, computer systems, and networking. One of the most interesting projects I had to do was to implement a Kalman filter (in C++ no less!) to guide a spaceship to the moon for my Artificial Intelligence class. It was also then (2017-ish) that I discovered the area of Computer Science which I wanted to specialize in.

I graduated in December 2017. I applied for Georgia Tech’s OMSCS program in early 2018 and had no issues getting in thanks to the GradDip program. By Fall 2018, I was working on my first OMSCS course.

Sidetracks

As I tried to transition from an accounting career to one in AI/ML, I tried to move laterally to a more technical role within my organization. Back then, machine learning/data science was still in its infancy and there were no such opportunities for me. So I tried to convince my Japanese company to adopt data-driven sales & operations planning (S&OP). Somehow along the way, I obtained a Master of Engineering degree from MIT, where I did a research thesis on using language models and deep learning to forecast prices. I am thankful to Georgia Tech’s OMSCS program, which probably factored into MIT’s decision of accepting me into their program. The modular nature of the OSMCS program also meant that I was able to take a break from mid-2019 to mid-2020 to complete MIT’s program first.

Reflections

Classes Taken

Through Fall 2018 to Fall 2022, I completed:

1. [Fall 2018] CSE 6242 Data and Visual Analytics
2. [Spring 2019] CS 7646 Machine Learning for Trading
3. [Fall 2020] CS 6475 Computational Photography
4. [Fall 2020] CS 7641 Machine Learning
5. [Spring 2021] CS 7642 Reinforcement Learning
6. [Summer 2021] CS 6601 Artificial Intelligence
7. [Fall 2021] CS 7643 Deep Learning
8. [Spring 2022] CS 7632 Game AI
9. [Spring 2022] IYSE 8803 Special Topics: High-Dimensional Data Analytics
10. [Fall 2022] CS 6515 Introduction to Graduate Algorithms

Memorable Moments

Some memorable moments of the program included:

• Draping sheets in my son’s bedroom to create a camera obscura.
• Fiddling with LaTex to fit 20-30 images into 8~10-pages reports for ML and RL classes.
• The anxiety to find groupmates and submit project milestones for group projects in DVA and DL and the relief when we found out that the grading was more than fair for our efforts.
• Trying to land a lunar lander using deep RL. I particularly liked this one because it felt like a continuation of the space mission from my undergraduate AI class.
• High-pitched peals of Minions’ laughter as they evaded dodgeballs for a project in Game AI.
• Working through an exam for HDDA at 3 am when the whole family was stricken with COVID.

Takeaways from OMSCS

Although I feel that OMSCS can be improved in several areas, it delivers incredible value in spite of being limited to an online format.

Through multiple very late nights, I developed many practical and theoretical skills. For example, I learned to work with various tools such as Docker and PyTorch, and how to track and version DL models using Weights and Biases. I also learned how to do backpropagation by hand, which I thoroughly enjoyed but probably belongs to the latter skill category.

My greatest takeaway is a habit of constant learning. After years of staying up late to finish assignments, I developed a habit of learning and working on my side projects in my free time, even after graduation; which I believe would be invaluable to my career.

Areas of Interest

Through OMSCS, I managed to identify areas where I would like to deepen my knowledge/expertise.

• Machine learning with graphs. (I love graphs! Thanks, Euler and the bridges in Königsberg!) This includes graph neural networks and their many variants.
• Bayesian machine learning, particularly familiarity with various MCMC frameworks and probabilistic programming languages such as PyMC and Pyro.
• Forecasting. This is mostly because I have been working with time series data all along. Recently with the growth in processing power and new deep learning models, there have been huge advances in areas such as NLP and computer vision. However, forecasting has yet to benefit much from them. I am quietly hopeful that we can see the same level of improvements in forecast accuracy soon.

I am also keen to learn about recent advances such as stable diffusion and generative AI. Luckily, OMSCS alumni can still register for classes even after graduation. I am truly grateful to Georgia Tech for creating such an inclusive and revolutionary program.

The Future

As of January 2023, there has been news of multiple layoffs by the MANGA/FAANG/WHATCHAMACALLIT companies. Given the explosive growth in computer science/bootcamp graduates in recent years, coupled with the advent of generative AI such as ChatGPT, some people might feel that a career in tech is bleak.

However, a common counterexample given is how Excel did not cause accountants to lose their jobs but helped to widen their job scopes. As a former accountant, I find this example apt and hilarious. I think that AI has the potential to free us from monotonous work and new jobs will arise from advances in AI. Should things turn south, I, for one, welcome our new AI overlords.

Having some free time now, I am looking forward to working and collaborating on awesome ML projects. I am always happy to jump on a call or connect over coffee to discuss interesting research or ideas, so feel free to reach out!

Go Jackets!