Rice University computer scientists Anshumali Shrivastava (right) and Aditya Desai created ROBE Array, a breakthrough low-memory technique for deep-learning recommendation models, a popular form of artificial intelligence that learns to make suggestions users will find relevant. (Photo by Jeff Fitlow/Rice University)

There’s exciting research coming out of Rice University in Houston that could put some of the most resource-intensive forms of artificial intelligence in the hands of small companies.

It centers on deep-learning recommendation models (DLRM), which are machine learning algorithms that learn from data, according to a news release from Rice. They help make suggestions that users find relevant. An example is a system that suggests products for shoppers based on data from past transactions, including search terms used, which products were offered and which were purchased.

“One way to improve the accuracy of recommendations is to sort training data into more categories. For example, rather than putting all shampoos in a single category, a company could create categories for men’s, women’s and children’s shampoos,” according to the release.

The best of these systems require more than 100 terabytes of memory and supercomputer-scale processing — only available to tech giants.

Rice’s “random offset block embedding array,” or ROBE, seeks to downsize those requirements — to a memory level equivalent to streaming music for about 1 ½ hours.

“Using just 100 megabytes of memory and a single GPU, we showed we could match the training times and double the inference efficiency of state-of-the-art DLRM training methods that require 100 gigabytes of memory and multiple processors,” said Anshumali Shrivastava, an associate professor of computer science at Rice.

Shrivastava is presenting the research at the Conference on Machine Learning and Systems (MLSys 2022) in Santa Clara this week along with ROBE Array co-creators Aditya Desai, a Rice graduate student in Shrivastava’s research group, and Li Chou, a former postdoctoral researcher at Rice who is now at West Texas A&M University.

“ROBE Array sets a new baseline for DLRM compression,” Shrivastava said. “And it brings DLRM within reach of average users who do not have access to the high-end hardware or the engineering expertise one needs to train models that are hundreds of terabytes in size.”

The ROBE Array research was supported by the National Science, the Air Force Office of Scientific Research, the Office of Naval Research, Intel and cloud computing company VMware.