Jesse Frumkin 700x550

Alumnus Dr. Jesse P. Frumkin Honored by Department of Commerce for Co-Developing Machine-Learning Algorithm to Analyze Patent Documents

Keck Graduate Institute (KGI) alumnus Dr. Jesse P. Frumkin, PhD ’12, was recently selected as a co-recipient of the Gold Award from the Department of Commerce. The award honors the distinguished scientific achievement displayed by Frumkin and his colleagues at the United States Patent & Trademark Office (USPTO)’s Office of the Chief Economist, Office of Policy and International Affairs, and Office of the Commissioner for Patents in developing a machine-learning (ML) algorithm to identify the volume, nature, and evolution of U.S. technologies directly from patent documents.

The group’s ML algorithm was found to perform substantially better than traditional query-based methods. Their methodology is now being imitated by other IP organizations and offices around the world and is poised to become the standard method in the field.

The award was motivated by the October 2020 report Inventing AI: Tracing the diffusion of artificial intelligence with U.S. patents, prepared by Frumkin and his colleagues. According to the report, annual artificial intelligence (AI) patent applications increased by more than 100% between 2002 and 2018, rising from 30,000 to more than 60,000 annually.

The purpose of the report was to gauge the potential impact of AI, examining patents to form conclusions regarding the nature and diffusion of AI technologies across a broad spectrum of technical areas, inventors, companies, and geographies. They utilized a machine learning AI algorithm to determine the volume, nature, and evolution of AI and its component USPTO technologies in U.S. patents from 1976 through 2018.

For patent applications and grants, they defined AI as comprising one or more of eight component technologies: knowledge processing, speech, AI hardware, evolutionary computation, natural language processing, machine learning, vision, and planning and control.

While most of the top 30 companies holding AI patents are in the information and communications technology sector (with notable exceptions such as General Electric, Boeing, and Bank of America), the application of AI is broad across the U.S., encompassing fields as diverse as medicine, fitness training/equipment, and agriculture.

Their methodology involved taking examples of patent-related documents and non-patent-related documents, modifying a pre-existing algorithm for those customized sets.

“In this way, we were able to create a score for each patent-related document to get a sense as to what extent it contains AI,” Frumkin said.

To validate their results, they manually assessed the scores to ensure that the documents were properly categorized by the machine learning algorithm.

“To me, the biggest takeaway is the team effort that led to that paper and the data set that resulted from it as well as the combination of using machine learning and validation in a governmental setting,” Frumkin said. “It’s important to be able to expand upon that for other types of technologies.”

He believes that patent searches are underutilized by academics.

“Being able to read public patent documents gives you a vision for the future of any given field or company,” Frumkin said.

At KGI, Frumkin earned his Doctor of Philosophy (PhD) in Computational and Systems Biology, where he did extensive literature review alongside his advisor, Professor Dr. Animesh Ray. In 2012, he started working for the USPTO in intellectual property (IP) law, a subject he had been interested in ever since taking business strategy classes with Henry E. Riggs Professor of Management Dr. Steven Casper.

After working at the USPTO for several years, he learned about the Office of the Chief Economist and realized that their research was like the type of work he had done at KGI.

“I knocked on the door of each economist in the office to introduce myself,” Frumkin said.

This led to a temporary position in the Office of the Chief Economist before becoming a Bioinformatics Patent Examiner. What he particularly enjoys about his job is the intersection of academics and government.

“One highlight was being interviewed to provide content for a Harvard Business School case study, which was actually one of my goals after taking Dr. Casper’s strategy classes,” Frumkin said.

Frumkin expresses gratitude to Ray for helping him to cultivate the intellectual curiosity that has defined his career trajectory.

“I remember asking him a question early on while working with him,” Frumkin said. “He said that he could tell me the answer, but he wanted to set me on the right trajectory. So he encouraged me to seek out the answer instead—something I have continued doing for the next two decades of my life.”

“He was also a very patient advisor. Even after I graduated, we continued to work together to get my PhD work published.”

Ray was impressed by Frumkin’s desire to expand his horizons.

“Jesse came with a degree in biology from MIT, where he briefly did undergraduate research in David Page’s lab, followed by a degree involving robotic vision from UPenn,” Ray said. “But somehow, he had managed to get his biology degree without doing laboratory bench work. All his research experience was using computers. But for his PhD, he wanted to do wet bench experiments. So, I thought, here’s a smart but crazy guy, who would probably burn himself and quickly give up doing bench work instead of computer work.”

At the time, KGI had a vibrant computational research group, of which Ray was a part. However, Frumkin was not to be deterred.

“He painstakingly mastered experimental genetics work, and finished his PhD with a flourish, mixing computing with experimental lab work,” Ray said. “Some of the published work from his PhD thesis, in my humble opinion, was ahead of its time. He is the kind of person I value immensely: thoughtful, imaginative, and in the kingdom of thoughts would go where no one has ever gone before.”

The views and comments expressed herein by Jesse Frumkin are solely his opinion, do not reflect the performance of duties in Jesse’s official capacity, and are not endorsed by, nor should be construed as, any viewpoint official or unofficial of the United States Patent and Trademark Office. Jesse confirms to the best of his knowledge that no information contained herein is privileged, confidential, or classified.