Deep Learner Spotlight: Christine Payne

Introduce yourself: What’s your background? Did you have a technical role prior to enrolling in the course?

Throughout my life I’ve juggled my love of both science and music. I was a physics major in college, then did a masters in classical music at Juilliard. After Juilliard, I continued on to study neuroscience and medicine at Stanford. Once I had kids however, I decided to continue only with music. For a few years, I was the pianist in a chamber music group with musicians from the SF Symphony.

Over time I missed being at the cutting edge of research and doing work that could directly help others, and I started looking for ways to return to my science roots. Then I discovered deep learning and AI. I love the way it has an impact on so many other fields (medicine, physics, and even music), and I knew I’d found the perfect field for me.

What sparked your interest in AI?

I first discovered deep learning by chance. As a pianist, I always need someone to turn pages for me, and I decided to create an automatic page turner using an eye tracker connected to my laptop. This worked really well, but I wanted to find a way to do this using only an iPad camera. When I started asking around for ideas, people pointed me in the direction of TensorFlow.  

At that point, I’d never heard of deep learning, and I looked for online textbooks and courses. I started a few, but didn’t get drawn in until I found the Deep Learning Specialization.  Once I saw the potential of deep learning, I was completely hooked.

What was your level of familiarity with AI before taking the course?

I was very much a beginner when I took this sequence. I had worked through Prof. Ng’s Machine Learning course on Coursera, but this sequence was my first introduction to Python, TensorFlow, and all the different neural net architectures. I did have a strong math background from my physics training, and that definitely made it easier for me to follow the courses.

What advice would you give to a learner who is just starting out?

Starting out in a new field is a bit like learning a new language. At first everything feels foreign and overwhelming, and it’s important to press on through that feeling. It’s ok if a video or a lab doesn’t make sense on the first pass, and it’s worth looking for blog posts or asking questions in the forum. As you circle back to topics, you’ll find that things start fitting into place.

Just like with learning a language, with programming you have to jump in and try things even if your code is full of mistakes at first. You can certainly learn a lot following the labs directly, but it’s even better to experiment and break things. Try starting with a fresh Jupyter Notebook and see if you can recreate the lab from scratch. Look for ways to improve the models and reach better accuracy, or take a model and apply it to a dataset you care about.  

Along the way, I’ve kept a notebook, and each day I try to write down one new thing I’ve learned that day. It can be a small detail or a big new topic, but it has to be something. In the beginning, this was helpful as a way of organizing what I was learning.  Now, it really pushes me to explore papers and to stay on top of a field that’s moving very quickly.

How do you continue to learn deep learning?

After I finished this Specialization, I worked through the courses and tried a few competitions on Kaggle. I also worked through Sergey Levine’s Deep RL Course. This past summer I created a music generator. It very loosely follows the jazz music lab in Course 5 of the Deep Learning Specialization, but I made the project much more elaborate. I was quite happy with some of the generations, and I have a webpage where you can guess if pieces are by human composers or by my generator.

What are you currently working on?

Right now, I’m a fellow on the technical staff at OpenAI working on multi-agent reinforcement learning.* The DL Specialization played a huge part in getting me here. In fact, as part of the fellows application, I had to answer questions about CNNs and RNNs and program using TensorFlow. A year ago I didn’t know about any of this.

Anything else you’d like to share?

After I finished the Deep Learning Specialization, I was invited to stay on as a Coursera mentor, answering student questions in the forum.  This turned out to be an invaluable experience. I thought I’d understood the course well from my first pass through, but answering other students’ questions helped me develop a real fluency. I’d highly recommend that everyone take advantage of the forums — pick someone else’s question and try to answer it. It’s ok if you don’t know the answer at first.  It’s really worth the time spent looking back at the lecture videos, playing around with the code in the labs, or searching online for ideas.

*As of May 2019, Christine is now a full-time Research Scientist at OpenAI who most recently published her work on MuseNet, a neural network that can create original music.

Want to build your own career in AI?

Get started by taking the Deep Learning Specialization!