How one teacher learned AI programming and lit a path for young coders
February 3, 2020
Erin Song, a science, technology, engineering, arts and math (STEAM) teacher in North Carolina, admits that less than a year ago she wasn’t entirely sure what the AI field of deep learning was or how it worked. But after a few months of online study, she packed up her 11-year-old son and their weekend bags and traveled to Northern California in August to compete in a 48-hour deep learning hackathon.
Her four-person team, which also included a computer science grad and a data scientist, took one of the top three awards with their platform, MineTorch. MineTorch teaches children how to use deep learning, the rapidly growing AI field that powers everything from speech recognition systems in smartphones to navigation systems in autonomous vehicles.
By the next evening, judges declared three winners out of the 24 submitted entries. First place went to learn2learn, a meta-learning library that makes reinforcement learning easier. Second place went to HelloWorldNet, a search engine that explores tens of thousands of stars to help hunt for new planets. And third place went to MineTorch.
That’s no small feat: There were 77 contestants — most of them AI researchers, machine learning engineers, computer science grad students and people who work in companies already applying machine learning — joining in the friendly competition. At the time of the event, which was hosted at Facebook’s Menlo Park, California, campus, Song had only three months of machine learning experience under her belt. “I was never focused on winning, because I’m a beginner,” she says. “This was just supposed to be a learning adventure. I wanted to show my son what humans are capable of doing.”
AI, machine learning, and deep learning are familiar terms to anyone who has experience with voice assistants, chatbots, email filtering, and other applications that learn to recognize complex patterns in data. But even savvy technology fans often find the field somewhat mysterious. Song says her hackathon success shows that with the right educational tools — along with plenty of focus, determination and help from a supportive team — it’s possible to master these concepts and apply them to real-world solutions.
The hackathon required participants — most of whom teamed up upon arrival — to create something that would have a positive impact on people or accelerate research using PyTorch, an open source deep learning framework. Researchers and developers at companies use PyTorch to build, train, and deploy machine learning models for applications in domains such as computer vision and natural language processing. PyTorch code is similar to programming in the popular Python programming language, which makes it easier to start building and training ML models.
The PyTorch hackathon participants were allotted two days, including the night, to work on their projects. Song, who is Korean American, says that when she arrived at the Menlo Park event early Thursday morning, she was blown away by the significance of the occasion. She, a working mother from the other side of the country who was accustomed to playing supporting roles at work and at home, was surrounded by brilliant technology minds from all over the world, and they all were being judged by the same standards: their ingenuity and the quality of their work. “It was so very exciting,” she says. She had a moment of nervousness, she adds, when she realized she was competing with professionals and full-time researchers who’ve worked in AI for many years. But she found the diversity of the participants exhilarating, inspiring and liberating.. “I felt like, wow, this is a place where I can be accepted and be a part of the community and not be held back by the way I look,” she says. “I wanted to savor every moment.”
That week, Song met, for the first time in real life, two classmates from Udacity (an online educational platform that teaches students skills in programming, data science, AI, and digital marketing). On a lark just a few weeks earlier, Song, Anna Unger (a blackjack dealer-turned data engineer from Stockholm), and Yu Sun (a Stanford University-educated consultant from San Francisco) decided to enter the hackathon together. The trio hadn’t yet decided what to create when they arrived at the hackathon, but that changed when another participant, an AI researcher from University of California, Berkeley, named Boyuan Chen, asked whether anyone wanted to help him create something for children. Boom: They had a mission and a fourth team member.
Overnight, they developed MineTorch. MineTorch is an easy-to-access, interactive user interface designed for children who want to explore machine learning. The drag-and-drop program helps users figure out how to explore and use data to build code on open source platforms. (MineTorch is available on GitHub here.)
“We couldn’t believe it!” Song says. “We wanted to create a product that anyone can use, because not every child has affluent parents or computer science parents. Every child, though, should have this accessibility and understand machine learning for the future.”
Song says she would have loved to have had access to MineTorch herself last spring. She was teaching math and science to her students of all ages, and thinking about how to better educate them and her son about future careers. She started reading articles about machine learning, but she didn’t really understand how it worked or how to even practically begin experimenting with it. Then she heard about the three-month PyTorch Scholarship Challenge from Facebook on Udacity. She applied and was one of several thousand applicants worldwide chosen for the program. For three months over the summer, Song studied the basics of developing deep learning models using PyTorch. During the program, she founded #sg_mom-is-wow, a Udacity study group where her peers, including machine learning newbie and future hackathon teammates Unger, could discuss parenthood, the challenges of being a mother who works in tech, and the role technology plays in their children’s futures.
“I was genuinely curious about how technology and AI is disrupting and changing the way we live and our children’s careers,” Song says. For her, just reading about machine learning didn’t cut it — she needed to experience it herself through practice, and Udacity taught her the ropes. What she didn’t realize was that the online class would also broaden her community. She says she met many people in her class from all over the world, and she says it struck her for the first time how lucky she was to be able to pursue learning and make a difference not only in her own career but also for kids who are just getting started. All her learning pointed her toward developing an educational tool for children who have the curiosity and desire to create.
And there are plenty of them. Millions of kids are already coding; in fact, Song and her team built MineTorch for children who want to participate in Hour of Code. By adding PyTorch to their coding, they have access to an open library of data to help them design games, code creative solutions for such things as setting a daily alarm, and even work together to brainstorm solutions to some of the world’s largest problems, such as climate change. “I think children are able to solve real-world problems,” Song says. “Adults get caught up in the bureaucracy, but children are so honest and have a very low tolerance for shams. I have more faith in children than in adults.”
Song says she hopes that she can connect MineTorch with teachers and that it will eventually be a tool in open learning education. “We’re at the beginning stage, but anyone can use this now,” she says. “Five years from now, I would love to hear that many people got their start on MineTorch.”
The next steps for Song are easy: She wants to continue learning, machine and otherwise. She is taking a class in software engineering and hopes to eventually get a job specializing in education technology. “I love looking at and paying attention to trends,” she says. “In five to 10 years, kids will be facing jobs that don’t exist right now. I’m just so happy to experience what’s out there and what’s possible.”
More information on Udacity's Intro to Deep Learning with PyTorch course is availablehere and here.
We're hiring AI scientists!
Help us drive scientific breakthroughs in core AI research