The machine learning maestro luring technologists to finance
Li Deng has been working with artificial intelligence and machine learning for nearly three decades. He spent 17 years at Microsoft and founded the tech company's deep learning center in Redmond, Washington. But for the past five years, Deng has worked in finance, and he's now on a mission to persuade other technologists to move to the sector too.
“In finance, the data is huge. In many cases, it’s much bigger than the data you’re dealing with in tech," Deng tells us. "Therefore, we need people who are very good at dealing with large datasets and who understand the essence of advanced learning models like deep neural networks.”
Deng spent 3.5 years at hedge fund Citadel, where he was head of machine learning until November 2020. In May 2022 he joined Vatic Investments, the automated liquidity provider set up by James Chiu of Jump Trading. At Vatic, he's slowly building a team, and not just with people from finance backgrounds. "You don't need to know about finance to apply machine learning techniques to finance data sets, says Deng. "Raw smarts are the most important character trait I look for. If people have a finance background, that’s good. But if not, it’s relatively easy to train people to apply AI techniques to finance problems.”
Deng says his network of AI contacts is considerable, and that he's still in contact with many of the 100 or so interns who went through his program at Microsoft. “I interviewed three people last week and I am interviewing four PhDs in the Seattle area tomorrow," he says. "I’ve been involved in the Seattle tech scene for a while and I know a lot of the really good engineers and researchers here. The good ones often contact me when they come available. They know that I can offer them challenging work.”
At Vatic, challenging work means using machine learning to generate automatic buy and sell signals based on learnings from as much data as possible, even when datasets are noisy. Most of the people Deng and Chiu interview are PhDs, but a handful only have bachelors degrees. “We have hired exceptionally strong undergraduates without PhDs, but generally we try to identify the top 1%,” Deng says. Chiu says PhDs are preferred because they demonstrate that people have used tools in, "a deep way that can be applied to our problems. The training of a PhD helps you both to define a problem and to understand how to solve a problem using the tools available.”
Although machine learning is excellent for generating buy and sell signals, Deng says human traders can still have the advantage: “Human traders can sometimes react faster and provide insights very quickly." By comparison, he says machines can make a mistaken assumption that the past will repeat itself in the near future and that this doesn't always hold, although techniques like smart temporal regularization can be applied to reduce this error.
Have a confidential story, tip, or comment you’d like to share? Contact: firstname.lastname@example.org in the first instance. Whatsapp/Signal/Telegram also available (Telegram: @SarahButcher)
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)