//php echo do_shortcode(‘[responsivevoice_button voice=”US English Male” buttontext=”Listen to Post”]’) ?>
— Part of an ongoing EE Times series: Diversity & Belonging in EE. Previously published in the series: Electro Soft CEO Karla Trotman, on Reaching the Top
In an exclusive interview with EE Times, Cerebras’ Natalia Vassilieva, detailed her life in the Soviet Union, unveiling her experiences as a woman in engineering that differs from what she encountered in the United States.
“Nobody pushed me toward [engineering], but that’s the environment I grew up in,” said Vassilieva, now senior director of product, machine learning at the Sunnyvale, Calif.-based artificial intelligence (AI) chip company.
Growing up in the Soviet Union, both Vassilieva’s parents were electrical engineers, and her grandmother worked as an engineer in steel manufacturing.
“One of the eye-openers for me when I moved from Russia to the United States was that in Russia, I didn’t feel like there was different treatment for boys and girls,” she said. “You would be encouraged to do what you liked to do, especially girls. When I moved to the States, I realized that it’s been a completely different environment, historically.”
Vassilieva describes how growing up, it was perfectly normal amongst her classmates for both parents to work, so she was shocked when she realized this hadn’t been the case for contemporaries growing up in the U.S. Equality between men and women in the workforce was a strong theme in Soviet Russia as far back as the early 20th century.
“[The Soviet Union] was not all negative, though you can argue about whether the reasons behind it were good or bad,” she said. “There were some good artifacts, and one is the way women were treated the same way as men, particularly after the Second World War.”
While North American women were encouraged to give up their positions when World War II ended to take care of their families, in Russia the war had a lasting effect. Such attitudes persist today.
For the young Vassilieva, early dreams of becoming a ballerina gradually gave way to an understanding that technical disciplines might be a better fit.
“Technical disciplines were clearer to me, and it was easier to understand what it means to do things right,” she said. “If you’re trying to solve a math problem, it’s clear and objective—you solved it or you didn’t. When you try to do a composition, it’s much more subjective how good your composition is.”
Vassilieva completed a Master’s in computer science and software engineering—which she said was mainly applied math—before a stroke of luck led her to the University of Grenoble, and later to switch the topic of her Ph.D. thesis.
“You follow in the moment what interests you, so I can’t say that I dreamed from the beginning that I would be exactly where I am right now,” she said, noting that AI didn’t exist in the same form then.
After her Master’s degree, Vassilieva’s French language teacher encouraged her to apply for an internship in France, and she picked the University of Grenoble since it had the closest focus to her background in applied math, and was close to the beauty of the French Alps.
“Only when I got there did I realise that the team I was joining was focused on the intersection of information retrieval and computer vision,” she said. “I got completely fascinated with the topic and when I got back from that internship, I changed my Ph.D. advisor and changed my Ph.D. thesis topic to time-based image retrieval.”
Computer vision at the time used simpler forms of machine learning (ML) for feature extraction, leading Vassilieva to pursue a career path in ML. She joined Hewlett Packard (HP) Labs, in St. Petersburg, Russia, to work on both computer vision and natural language processing in 2007 and spent 11 years there before joining AI chip startup Cerebras in 2019.
At HP Labs, Vassilieva led a team tasked with identifying the compute requirements for different analytical workloads, including deep learning (DL), Monte Carlo simulations, graph inference, and many other types of data analysis and analytical algorithms.
“Through that work, I formed an opinion on what would be good hardware architecture, or what properties do I want to have in the hardware to be efficient for deep learning,” she said.
While algorithms have since evolved significantly, at the time, the biggest problems were training with large batch sizes and distribution of training across many GPUs—problems which are only partially solved today.
Cerebras’ wafer-scale architecture matched Vassilieva’s vision for what properties an efficient DL hardware architecture should have, but joining an early-stage startup was quite different to working in a large global organization like HP.
“It was interesting to join a company which has one mission—where everybody speaks the same language and has the same goal,” she said. “It’s dynamic and exciting, you get to wear a lot of different hats so it means more freedom, but the main reason was that the whole team is driven by one single goal, and you can’t say that for huge enterprises.”
Vassilieva, 43, said she doesn’t see widespread, universal challenges for women in the semiconductor industry today.
“I guess I am lucky to be accepted for who I am, but [at this stage of my career] I don’t find it hard to say openly what I think, I’m not intimidated by high-ranking men, and I believe open communication is the most valuable, most direct way to get things done,” she said.
While she admits this kind of self-confidence would have been harder to find as a younger person, she advises girls and women that confidence is still key, as hard as it may be.
“Overall, I observe that boys and young men are more persistent and more self-confident, on average,” she said. “I didn’t know how to do that [earlier in my career], but [women] need to treat [themselves] as equal…. We need to believe we can do a lot of things, and maybe to trust ourselves more and be more confident in what we do.”
