Empowerly

Student profile

It’s 6pm, and my kitchen is a mess of flour, eggs and cabbage. I had seen videos of Okonomiyaki -- Japanese savory pancake -- and was determined to make it myself. I called friends over, and though we started with a recipe, cooking quickly devolved into throwing whatever seemed tasty into the batter. Fresh off the grill, the pancakes were mouthwatering. As huge fan of Japanese cooking, I’ve spent many afternoons searching for sauces at the local market, or experimenting with new recipes at home. I’m left with delicious food, and a great excuse to gather family or friends.

Although you may not yet know what you want to major in, which department or program at MIT appeals to you and why? (100 words or fewer)

Since I was small, automation has been my “holy grail”. I’d write autoclickers to get ahead in computer games, or bots to gain levels on forums. The day I first heard of artificial intelligence -- the ultimate automata -- I couldn’t sleep. I dove headfirst into online courses, from image generation to robotic movement. The field of AI is still primitive, and I’m determined to leave the next big mark. As an intern at OpenAI, I experienced the research life, and I aim to continue. At MIT, my eyes are on CSAIL, especially the UROP programs.

At MIT, we bring people together to better the lives of others. MIT students work to improve their communities in different ways, from tackling the world’s biggest challenges to being a good friend. Describe one way in which you have contributed to your community, whether in your family, the classroom, your neighborhood, etc. (200-250 words)

The hackathon. A mass of people armed with technology, split into teams, and given 24 hours to bring their craziest ideas to life. I've been hooked since middle school, and by the end of freshman year, I’d participated in over 15 of the events. But my teammates at our school’s 5-man Engineering club were graduating, and suddenly, only I was left behind. I took things into my own hands: to revive the club from a single member to an entire roomful.

Many students were interested, but unconfident in their skills. So I recruited an officer team, and designed workshops for beginners. That year, Engineering grew to over 30 dedicated members. Together, time flew by fast-- from flying to hackathons across the nation, to teaching our robotic arm to draw. Building a quadcopter only to crash it into the curb wasn’t a failure, but a great story for the next meeting. With the team, I’ve organized a hackathon at our school for three years, attracting over 200 locals.

Entering senior year, I’m taking a different approach. I’ve let new faces take over the leadership, to prepare for when it’s my turn to leave. Last month, I brought a team (2 first-timers!) to PennApps, the nation’s largest college hackathon. But rather than programming myself, I acted as a mentor. Overnight, the two first-timers built their first iPhone app, and after 24 hours they emerged from the judging hall -- with the top category award. Seeing the looks on their faces, I couldn’t help but smile along.

Describe the world you come from; for example, your family, clubs, school, community, city, or town. How has that world shaped your dreams and aspirations? (200-250 words)

"This is amazing." "How do you beat the first boss?" When I first uploaded my Flash games to Kongregate, I wasn't expecting the comments. Honestly, I had just wanted to earn some pocket money from advertisements. But seeing requests like "more levels..." made me realize that behind the view-counts and usernames, there were real people, with their own ideas and opinions. Replying "here's the sequel!", I grew to appreciate the online community of game enthusiasts. A month later, I was auto-banned for being under 13 years of age. I had to move to other platforms, but I kept releasing things -- not for the money, but for the people out there who enjoyed them.

Even today, this mindset's leading me to new places. I've got this dream, partially inspired from fantasy games of robot races: a computer as intelligent as humans are. A virtual mind to talk, create, and laugh like we do. It's a distant vision, but I'm not alone -- there's an entire field of people working towards computer intelligence. Of course, everyone starts as a beginner, and it's the helping hands of the community that help them thrive. I've started a journal, where I break down my own projects, along with interesting research, into easy-to-understand terms. When publishing my first paper, I put focus on releasing open-source code and non-technical explanations, to help out beginners. "Best tutorial I have come across ... after struggling last ~3 weeks". To this day, it's the people behind comments like this that make everything worthwhile.

Tell us about the most significant challenge you’ve faced or something important that didn’t go according to plan. How did you manage the situation? (200-250 words)

When artists are struck by inspiration, they rush to the paint board. For me, I make computer games. And in 9th grade, I started my most ambitious project: Skyflower, a shoot-em-up adventure for the PC. Every time I looked over the plans, only one thought came to mind: "this is gonna be awesome". For six months, it was straight to the laptop after school. From online tutorials, I learned to draw, design levels, program enemies, and market - it was a one man show.

During an afternoon math class, I received amazing news - Skyflower had been accepted into Steam, the top marketplace for computer games. But through the excitement, things started to fall apart. Without anyone to help me, I couldn’t keep up with my own schedule, and buyers began complaining about the lack of updates. My attempt at a multiplayer mode failed and left a sea of negative reviews.

Honestly, I was devastated for a few weeks. I felt if I had planned ahead better, Skyflower could have made it big. Even so, I'm accepting those mistakes, and moving onwards. I've got a game I can proudly call my own, and the skillset from making it a reality. Today, I'm working towards a new game -- a fast-paced platformer centered around dueling giant bosses. The goals are still just as lofty, but I’m keeping organized, pacing myself, and most importantly, I've recruited a team to back me up!

Please tell us more about your cultural background and identity in the space below (100 word limit). If you need more than 100 words, please use the Optional section on Part 2.

My parents and I are from Indonesia, specifically the small cities of Tegal and Kebumen. We are the first from our extended family to come to America. In our house, we often host relatives and family friends from Indonesia for extended periods of time, so I’m commonly living with 7 or 8 people.

Activities

AI projects and research

Independently designed and built projects such as automatic sketch colorizer and chat simulator. Released two research papers. Maintain a research blog attracting ~6000 readers a month, along with multiple open-source projects with 150+ branches from others.

Hackathons

Competed at 20+ college level hackathons around the country. Brainstorm ideas, manage work with teammates, and present with confidence and a smile. Led school teams to compete, as 2-year president of Engineering club. Organized school’s annual hackathon three times.

Tae Kwon Do

Practiced for 10 years, currently at the black belt level. Train in sparring and forms, both unhanded and using wooden weapons. Lead class exercises as an assistant instructor, and teach lower level belts in forms and techniques. 

Game Development

Created games from many genres, by planning levels, drawing art, programming game logic, and marketing. Launched a paid game on Steam with ~21,000 purchases. Released various smaller titles to the App Store and browser websites. 

Provide a brief description of your research.* (1500)

This research deals with the learning aspect of artificial intelligence. An idealistic goal is to develop a computer agent that can learn to solve any arbitrary task at a human level. A concrete example is teaching a robotic arm to pick up objects. We want agents to learn strong strategies - in this case, a function mapping from the arm’s position to the movement on its joints.

State-of-the-art methods represent strategies as neural networks, which are trained through reinforcement learning. Agents iteratively improve by experimenting many times, then using the collected information to perform better next time.

In this research, we claim that strategies can be broken down into parts which are often shared between multiple tasks. For example, a robotic arm could learn to move or rotate its hand in three directions - primitives useful not only for picking up objects, but also for drawing pictures or solving a Rubik’s cube. It is much easier to learn a solution over a hand-movement based abstraction than to deal with individual joints. As such, we present a method of learning a decomposition of hard tasks into smaller, easy-to-handle ones. We surpass baselines, and successfully teach humanoid robots to walk and crawl under obstacles, and ant-like robots to complete obstacle courses. 

At a high level, this work brings together the growing fields of meta-learning (learning to learn faster) and hierarchical learning (breaking down tasks into simpler parts).

Summarize your contributions to this project. What were the specific tasks that you were involved in?* (500)

This research was my summer-long project during my internship at OpenAI. I came up with the idea, with guidance from my mentor. Aside from the physics simulations, I wrote all of the code for this project. I also performed all of the experiments. The bulk of the research paper was written by me. As this was my first paper, two others at OpenAI provided feedback on the later iterations. I also created the supplementary videos and wrote the public-facing blog post featured on the OpenAI site. 

What did you learn about the research process and your research interests? (1500)

The hardest part of research, by far, was coming up with “The Idea” -- what problems are interesting, and how can they be solved? It took four weeks of primitive experiments before I could come up with a project proposal that both my mentor and I were happy with. A piece of advice I heard over lunch greatly shifted my mindset: instead of aiming to improve performance by 20%, aim for 2000%. Rather than tackling low-hanging fruit, I began to think about the big picture, as a 2000% gain wasn’t going to arise from just small modifications. 

Once I had The Idea, I began implementing the algorithm and performing experiments. Molding the research into a presentable, final piece was a test of endurance. I learned to structure experiments as sets of specific hypotheses, rather than presenting arbitrary findings. I also got a taste of the revision process. Working on a single paper for a month was at times draining, but it was incredibly rewarding when I got to share code and publish my ideas to everyone.

On a higher level, this research project has only convinced me more that artificial intelligence will be the future, and I want to be a part of it. The field is huge - from computer vision, to voice synthesis, to robotics. Yet, everyone is fixated on essentially the same dream - a computer as smart as humans. As such, the various fields of AI have come together, building on each other’s work. With progress at such fast speeds, it’s exhilarating being a part of the movement.

What do you make?* (900)

Describe for us the kind(s) of thing(s) you make. This should be a general overview of your project(s); you will be able to submit pictures and video with specific descriptions later. Often, the most helpful explanations for Maker Portfolios emerge in the captions to the portfolios, where they can form a kind of narrative to help us understand how a project or projects unfolded.

Whenever a burst of inspiration comes, I make games. They are always centered around a single experience that I want to convey to the player, like the thrill of dueling a giant boss with only a single sword, or the beauty of dodging patterns of 1,000 bullets. Everything else -- storyline, art, gameplay, sound -- revolves around that vision.

On the hardware level, I’ve spent many hours tinkering with “robots” made from Arduinos, servos, cardboard, and duct tape. My smartphone usually ends up taped on as an all-in-one sensor. These break all the time, but rebuilding from random parts is a major part of the fun.

More recently, I’m embracing the idea of using my computer as a brain -- to procedurally create fantasy worlds, auto-control robots, or draw never-before-seen pictures. When the programs do something completely unexpected, it really feels like they’ve come to life.

How do you make?* (900) Describe your making process. What tools, techniques, or programs do you use? Which guides, tutorials, or support communities help you? Where do you make things (at home, at school, in a community space, on your computer, etc)?

I almost always start on pen and paper. When an idea comes, I’ll quickly draw out some designs. The most important thing is the end goal -- what I’m trying to make. Once I’ve got that down, I can think about how to actually make it happen.

The tools or techniques I use depend on necessity. It’s how I’ve ended up with most of my skills. I wanted to build a multiplayer game, so I learned server-side Javascript; or I wanted to make diagrams, so I learned how to use Photoshop. Google’s my number one guide to online information (tutorials, code, forums) -- I’ve become a jack-of-all-trades, at least enough to know what to look up next. 

My house has become the go-to gathering place for my close group of friends, so most of what I build is done here. We also go to hackathons -- an event where teams have 24 hours to build a project. Some of the best ideas have arisen from 3-am discussions there.

What's the most meaningful thing you've made?* (900) Describe for us the single most significant (to you) thing you've made and why it holds such importance. You might tell us about what challenges you encountered while making this, and how you went about solving these issues. This is an opportunity for you to help us better understand you as a maker.

My room is full with stacks of quick ideas I’ve sketched up. I’ve always wanted to build a computer program that could turn my messy sketches into clean artwork. A few months ago, I took my first steps towards this goal, and built Deepcolor -- the automatic sketch colorizer.

To train the core neural network, I needed a big dataset of sketches, and a powerful computer -- none of which I had. Instead, I wrote a script to scrape popular image hosting sites. I also persuaded NVIDIA’s university relations dept. to sponsor me a graphics card. From then, it was over a month of experimenting with network structures and training strategies. When the program finally generated likeable artwork, I had the greatest grin. I released a demo online, along with code and a post-mortem. Turns out, there are others working towards the auto-drawer dream -- and it felt great having them build off my work.