HP: How might HP use the algorithms you’ve been working on?
We’re looking to identify when you are over-working to achieve a goal. Say you’re learning to be a pilot and training in a VR flight simulator. If we can identify when you're in a situation where you are likely to make a mistake, we can make your trainer aware that you are in that state and also adjust the simulation to help you learn better. So the idea is to create a more personalized learning curve for specialized situations where you require somebody to be confident going into that field.
HP: How has your project changed as it has developed?
What’s changing is that we’ve decided we want to add some new biometric signals into the model that we are building. I currently have a kind of meta-model that takes a number of separate AI models that are each trained based on a particular signal and blends them together to make an algorithm that is more robust and reliable than any of them on their own. But now I want to take at least one new signal and train a new model for it – and then, once I have it working, add that to the meta-model.
HP: Is that meta-model something that you’ve created yourself?
People have created these kinds of “models-of-models” before, but what’s new is that I’ve built one that is optimized for our particular situation. Another important part of what we’re trying to build is that we want it to be modular. We want to be able to plug in new sub-models based on new kinds of biometric signals as we develop sensors that can measure them.
HP: What's the big challenge in doing this?
I think the biggest problem we face is distinguishing signals from the noise around them. People in the field have tended to develop their algorithms based on really clear data. But we’re interested in being able to make sense of signals where the data isn’t so clean – where you are using more affordable sensing technologies to gather your data inputs.
HP: You interned at HP Labs last summer too, right?
Yes. Last year I was working on generative modeling – creating a way for people to make digital objects even if they aren’t skilled designers. The idea was to allow someone to design sunglasses using natural language or rough sketches, say, and the model would translate the high level design intent into a sunglasses design suitable for 3D printing. We submitted one patent and we got a paper out of it, too. I'm hoping to do both of those things this year as well.
HP: Have your two HP Labs internships influenced your thinking about what you’ll do next?
They have definitely made me want to go to grad school, which wasn't something I was sure about before. I also now know that I don't want to end up doing just software engineering as many computer scientists do – I want to do research and to continue inventing new things.