It’s not a simple thing to get a car to see what we see.
“The world is very complex. That’s what makes vision for self-driving cars a challenge. There are millions of scenarios and millions of contexts,” says Carolina Parada (’04, ’06 MS Elec. Eng.) from her home in Boulder, Colorado.
A senior manager for Nvidia, a company probably best known in the video gaming community for its top-shelf graphics cards but with a strong presence in the machine learning market, Parada and her team are working on machine perception, a key piece of getting self-driving cars safely on the
Parada is no stranger to the cutting edge of machine learning. When you say “OK, Google” to your Android device, you’re using a technology she helped develop. Now she’s got her sights set on teaching cars to see.
Many of the things we take for granted, like being able to tell a leafless street tree from a skinny teen standing on the edge of the sidewalk, are not so easy for computers. And once it does learn the difference, a process that involves showing the computer “many, many examples” of both trees and standing people, says Parada, it has to learn that difference in a vast array of contexts, from rain and fog to night and day. And that’s before we even get into road signs and markings, which vary by region and country.
It’s “deep learning” that “enables a computer to learn from vast amounts of data,” she says.
One of the ways that Parada and her team got voice recognition working was by “dogfooding.” As in, eat your own dog food, because you’re then going to be incentivized to get it right before you unleash it on millions of customers. She’d test the system at home, asking her two daughters to interact with it.
“The system learns what you give it,” she says. “If the data is more male voices than female, then you’re fine-tuning the system to work better with male voices.”
Not surprisingly, there are lots of male voices in the high tech industry, one reason Parada values diversity on her teams. “It can only help,” she says, to have multiple perspectives on a problem. Which pretty much sums up her own perspective on engineering intelligent machines: fail fast and then try something different.
Parada certainly hasn’t been afraid to take risks. Originally from Venezuela, she was in engineering school in Caracas but dreaming of living in the United States. Her dream was realized when she and her husband, Jorge Bernate (’04, ’06 MS Chem. Eng.), got accepted to Washington State University and moved to Pullman.
Parada and her husband also had their two daughters while studying at WSU. “I had the girls during the breaks,” she says, laughing. “One in summer, one at winter break.”
She recalls how supportive everyone was. “People were always offering to help. I opened the door one day, and found a huge present, a bunch of baby clothes, hand-me-downs from neighbors.”
One of her engineering professors, Shira Broschat, is still a friend. Says Parada, “Shira was in Boulder a couple months ago, and she and her husband came over for dinner.”
Parada worked for Broschat for a couple years on computer simulations for bioengineering as well as electromagnetics. And when Parada decided to go on for a master’s degree, she jumped into robotic control systems.
“The mindset of taking risks is critical to the mindset of being an engineer,” Parada says. The mother of two girls adds, “Girls need to be taught to not be afraid to fail and take risks, to push their limits and maybe be a little uncomfortable. Most of the time you’re going to be surprised, because it’s going to work out and you’ll build your confidence over time.”
Parada regularly makes the lists of top women and Latinas in tech, so when she says that a passion for experimentation and exploration are keys to success, she’s speaking from experience. Just try it, she says: “This may fail but I’m going to try anyway because I think it’s worth a shot. That’s how you change the world, by trying things you haven’t tried before.”