Eric Allen: CS + Deep Learning

Eric Allen, CS Ph.D. alumnus and Deep Learning Researcher at Two SigmaEric Allen, now a Senior Vice President and Deep Learning Researcher at Two Sigma, flew to Houston during his senior year at Cornell University to explore Rice’s graduate program in computer science.

“That spring I was accepted at Rice, and they flew down prospective graduate students. I got to meet a lot of the CS professors, and I remember Keith Cooper was talking about Duncan Hall, which didn’t exist yet. It was still was being built.”

He discovered many potential mentors on that trip, including his future Ph.D. adviser, Corky Cartwright. Allen said, “Corky and I had lunch at Goode Co and we really clicked. I also met Moshe, Devika, and Matthias – it was quite an energetic culture and I could tell I would learn a lot here, that it was a place where I could really thrive.

“The warm weather and Texas barbecue didn’t hurt either. I’d grown up in Buffalo, NY and when I went to Cornell, I found that Ithaca, while beautiful in the summer, was actually colder than Buffalo in the winter. We joked that living in Ithaca in the winter must be similar to living on Mars.”

Allen said he was shaped not only by the faculty members, but also by other CS graduate students. “Shriram Kirshnamurthi had a big influence on me. He’s now a CS professor at Brown. My Ph.D. program ended up being a critically important part of everything I’ve done since I left Rice. The lessons I learned from Corky, Matthias, Moshe, and Devika – which sometimes I wasn’t ready to learn – played a big role later in my career.”

He also met industry representatives that shaped his career. While a Rice graduate student, Allen met Steve Heller, who was the Senior Research Director at Sun Microsystems and visited Rice during a corporate affiliates event in the CS department. He worked with Heller at Sun Microsystems and they are now colleagues at Two Sigma. Allen also met Guy Steele, a Sun Fellow at the same company, through Rice.

Allen said, “Guy is one of the inventors of Java and Scheme, and he was one of my heroes even when I was an undergrad at Cornell. Corky had taken a sabbatical and worked at Sun for a year and wrote a paper with Guy. When he came back to Rice, he introduced me to Guy, who ended up becoming my key mentor at Sun for years. I can’t imagine what my career would have been like if I had not come to Rice.”

Artificial intelligence drew Allen into pursuing a Ph.D. and it continues to drive his research. He said he studied knowledge representation and formal methods, so it was natural to transition to studying type systems. “There are deep connections between type systems and formal logic,” said Allen. “Devika tried to convince me in the 1990s that neural networks would take off but I couldn’t believe it. I wanted to focus on symbolic approaches. She was really ahead of her time.”

When he was hired in 2003, Sun had just won a DARPA high performance computing (HPC) contract to build a “petascale” computer (several orders of magnitude more powerful than the highest performing computers at that time). Allen said, “I was focused on type systems, new languages for massively parallel computing, and massive data sets on very powerful machines. And formal methods for understanding these systems. I was asked to join the Sun team that was tasked with coming up with a language to determine how to express computations at that scale.”

Sun’s biggest rival in HPC solutions was IBM. Allen said he ended up presenting at the same conferences as the head of the competing project at IBM, Vivek Sarkar. “We knew each other because we’d be talking to the same audiences about our company’s latest innovations.

“I’ve always loved teaching. When I arrived in Houston at Two Sigma, I reached out to Rice, looking for ways to collaborate. Vivek was the chair of the CS department, and it was quite an opportunity to co-teach with him for several years.”

Before joining Two Sigma, Allen said the advent of big data renewed his interest in methods other than artificial intelligence and symbolic reasoning. “Some of the best big data environments to work with are in ecommerce because their datasets are enormous,” he said. So he left Sun and became a data science engineering manager in Austin, first at BazaarVoice and then at RetailMeNot.

Allen said, “Two Sigma also falls into this category — we are all about data. Our mission is to “discover value in the world’s data” and my role is increasingly to focus on research in deep learning, which relies critically on big data. In a sense, I came full circle. I was convinced that neural networks would not catch up and now they’re all I think about.”

Since 2001, Two Sigma has been blending principles of investment with principles of technology and innovation. The fifty-billion-dollar investment manager is among the top five such institutions in the world. Through its investors, Two Sigma helps support the retirements of millions of people around the world, fund breakthrough research and education and help a wide range of charities and foundations to grow and thrive.

Diving deep into Two Sigma’s enormous datasets is now Allen’s passion. “A good day is when you find a new way to think about a problem, to connect and apply technology with a novel approach that has an impact.”

That kind of passion and innate curiosity about finding ways to impact the work and lives of others kept Allen grounded in graduate school. Looking back at his Ph.D. experience, he says graduate school is not for everyone. “Coming out of Cornell, I was young and immature in a lot of ways. Grad school is a humbling experience. It’s a crucible. It takes your brashness and replaces it with knowledge.

“But it is hard to get the kind of training you get in a Ph.D. program anywhere else. It teaches you how work on an ambiguous problem over a long time, with little feedback. There is no fixed time to find the solution – you get to leave with a degree once you’ve solved the problem and completed the work.”

Allen recommends graduate school to anyone who wants to deeply understand a body of knowledge. He said, “The single best decision I’ve made in my career is getting my Ph.D. The success I’ve had in my career, I can point back to lessons I learned from being a grad student at Rice.”

He said the Ph.D. training is especially important for anyone who wants to be a researcher. “In CS, if you want to push technology forward, to invent technology that doesn’t exist, you need to be able to build on an existing body of knowledge and contribute to it competently. Our field is constantly changing, so it’s a tremendous advantage to be able to contribute at that level.”

His advice to prospective and current graduate students mirrors advice he received from Moshe Vardi. “A lot of advice that Moshe gave me centered on learning to focus on a particular area and take it to completion. So stick with it. Persistence is the key. Learning to be persistent has benefited me more than any other strength. Many times at Rice, I entertained the idea of leaving and I’m really glad I finished it, took it to the end.”

Eric Allen’s adviser was Corky Cartwright; he completed his Ph.D. in 2003.