Original Source Here
How Hacking and AI Research are Related
Why being a great hackathon contestant can help with a research career
At first, Hackathons and AI research seem like they have few similarities, mainly the fact they are both CS and AI-related. However, after being involved in both for over a year now, I’ve learned to use hackathons to improve my research skills and vice versa (similar to how multi-task learning works). While research is a real occupation as opposed to competing in hackathons, there are lots of portable skills between the two that are important to highlight. Both of these have basically become my life outside of high school now, so I’ve been deeply analyzing how the two are similar and why I seem to enjoy both so much.
My inspiration for this article comes from recent news. This week, it was announced that I was selected for Major League Hacking’s Top 50 Hackers of 2021, which is an honor for the most successful and impactful hackers in the hackathon community. In my profile, I was asked to describe what I want to do in the future, and when I wrote about writing about research, I realized how many skills hackathons have taught me that are transferable to AI research. In this article, I will talk about both hard and soft skills that are shared between the two and why they are important to be a successful researcher.
Obviously, hackathons and research share a lot of hard skills from a programming standpoint. Most of my hackathon projects are based on AI, and my role on my team is to build and train AI models. Therefore, hackathons are great practice for using frameworks such as PyTorch. I also have more opportunities in hackathons to practice data processing and other skills essential to coding AI models for research.
One important difference is that for hackathons, I often find myself using Tensorflow since the models are much more simple and don’t have to be high-performance. Tensorflow is much more beginner-friendly in my opinion, and when I don’t have to write complex training loops, Tensorflow does the job much more efficiently. However, for research, writing advanced training loops and building complex models is required, and therefore the versatility and technical advantages that PyTorch provides makes it my favorite ML framework. I, however, won’t be able to give the best breakdown on which is better, so read this article instead.
Another hard skill that hackathons have taught me is presentation and writing skills. An often overlooked fact about hackathons is that the key to winning is to write an effective report about it and present it correctly. The project can be the most technically advanced project at the event, but if you don’t have the presentation skills to portray it properly, you will not win. Similarly, for research, the research you conduct is only as good as how you present it. In the paper, you must do a proper literature review, explain the method and all related terms, and present the results in a digestible way. My most recent paper actually struggled with some of these due to the limited time we had to complete it. While the reviewers praised the novelty of our method, because it was not presented very clearly, it gained negative reviews. Hackathons have helped me develop presentation skills not only for lay people but also for technical referees, overall making me a stronger researcher.
From a hard skill standpoint, the best thing hackathons have taught me in relation to research is how AI works in a variety of fields, such as NLP, robotics, and audio. I first learned NLP through a hackathon project, as experimenting with code and data over a low-tasks weekend allowed me to understand the basics of NLP and explore its potential applications. Since hackathons are just one weekend, they allow you to play around in a new field with little risk but learn a lot at the same time. This last weekend, I decided to explore audio nueral networks, and in the process, I read many research papers on audio processing with AI, allowing me to determine if it would be a potential area of research for me in the future. The free-flowing nature of hackathons helps tremendously with research endeavors.
Aside from the ones listed, there are many other obvious hard skills that hackathons and research share, but these are the ones I believe are notable. More important, however, are the soft skills that hackathons share with research.
The first soft skill I want to discuss is being able to work under pressure. Hackathons are built to induce pressure. 24–48 hours is very little, so to build a full prototype application requires amazing efficiency and lots of quick thinking. To be a great hacker, you must be able to code quickly, build MVPs, and recover from setbacks fast. In research, there are also hard deadlines that must be met. While a conference submission might take 3–4 months to complete, when compared to how much work is required for a full research paper, it’s a similar intensity on a weekly basis to a hackathon project. Therefore, you must also be able to code quickly, figure out how to train lots of models at the same time if you don’t have a fully-specced out deep learning rig, and write the paper in a timely manner. Doing hackathons certainly helped me with time management in my first few research projects, and finding an activity where you are so time-constrained like hackathons is rare, meaning its impact on your efficiency is unique.
Another important skill is being able to recover from mistakes. As expected, there are always failures and mistakes, but being able to recover from those mistakes is a huge skill that I have learned from hackathons. Being able to reconstruct miswritten code or retrain a model because certain parameters were not set correctly are important abilities that I have learned, and they are very important in the research field, as great precision and attention are required for running proper experiments and creating reproducible code. It also teaches teamwork and communication, as both research and hackathons require you to work with others on a frequent basis. Work for a project is always split between multiple people, and in order to be successful, you have to be a good communicator and know your role on a team.
Most importantly, however, hackathons have helped me develop a creative and innovative mind. Each week, hackathons force me to formulate a new project that is novel to the field, as uniqueness is a massive criterion in judging. However, the ideas most often can’t be outlandish, but need to be real solutions that are helpful to people and improvements from current solutions. Similarly in AI research, each paper presents a new method of achieving a task that often is an improvement of previous work, and lucky for me, hackathons have shifted the way I think, allowing me to hopefully be a great AI researcher in my future career.
Honestly, a lot of these soft skills can be found in other activities or jobs, but paired with the hard CS and AI skills that hackathons provide, being a hacker certainly can positively influence a research career. The soft skills are a great addition.
Final Thoughts and My Career Plans
Hackathons have given me the software engineering skills to hopefully join a research lab in the next few years once I enter college. Both the hard and soft skills that I’ve learned from hackathons have helped me in my research career so far, and I expect it to continue to help me once I enter college and become even more involved in research.
A major dilemma I have thought about is whether I want to be at an institute such as Google in their AI or DeepMind Labs or a university setting. It seems that corporations such as Google or Facebook are significantly harder to be employed by, but in the computer vision field, lots of the top breakthroughs are from corporations, which I would love to be involved in. However, university research seems to be more interconnected with other disciplines and has greater collaboration and impact, which is also an important factor to consider. Currently, I am an intern at Stanford Medicine in the Radiological Sciences Lab, where I have published multiple papers applying computer vision techniques to medical imaging tasks. The application of my knowledge to specific tasks such as denoising or image quality assessment has a much clearer outlook on its potential and impact, but the theoretical approach that general computer vision research is centered around certainly captivates me. While I have lots of time to decide, the sooner I decide the easier my career will be.
In the next few years, I want to narrow down to a single area of research in a single field of research. Currently, I conduct research in two main fields: medical imaging and general computer vision. Depending on my success in these fields as well as my future opportunities, I see myself being more involved in industry research and general computer vision, as it appeals to me more. My current background and interests indicate I am more suited for the general computer vision field, but medical imaging is still compelling to me, so if I learned more about the biological background of my work, I think I would be more interested than I already am. Nonetheless, I am excited about my college experience next year and hopefully, I continue attending hackathons concurrently with being involved in research so that I continue to improve at both. If you all have any thoughts or advice on this, let me know! Thanks for reading.
Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot