“It’s been a brilliant experience. The students have pushed us to learn and better understand more things.” Pavan Kapanipathi, IBM research scientist, had this to say about his participation in the CDS Industry Mentorship program. The program, an exclusive benefit for members of the CDS Industry Affiliates Program, matches small teams of data science Master’s students with an industry-proposed project. Each team works under the guidance of an industry mentor and a UMass PhD student mentor, in a course supervised by CDS faculty director Andrew McCallum. The Spring 2020 cohort wrapped up in May, with over 70 students on 22 projects from Amazon, Chan Zuckerberg Initiative, Facebook, GE Healthcare, Goldman Sachs, IBM, Lexalytics, Microsoft, Oracle, Raytheon, Stanley Black & Decker, the UMass Classics department, and Voya Financial.
Kapanipathi and his colleague Kartik Talamadupula mentored the team working on the IBM project Evaluating Natural Language Inference Models for Question Answering. Natural Language Inference (NLI) is the task of determining whether a hypothesis is true, false, or neutral, and has become a fundamental task in NLP (natural language processing). The goal of this project was to evaluate use of NLI models for downstream applications like question answering (systems that automatically answer questions posed by humans in natural language). The team of students, along with their industry and UMass mentors, authored a paper based on their work, which was submitted to EMNLP (Empirical Methods in Natural Language Processing), the world’s premier NLP conference. “The paper was a very strong submission,” Kapanipathi said. “The students put in so much work, and that motivated us to find time we didn’t know we had. It was almost shocking for me that students can work as hard and at the level that they did.” Even though the program ended in May, the group is continuing to collaborate. “The best collaborations tend to linger on after the course has ended,” according to Kapanipahi. “We are currently writing a follow-up paper, which these students are working on while doing full-time internships this summer. This is close to the best-case outcome in terms of collaboration -- we want to continue the work purely on its technical merits.”
This is the third year that Kapanipthi and Talamadupula have participated in the program, and each year, they find immense value in working with students on an IBM idea or issue. “Sometimes there are ideas we want to pursue that don’t completely align with IBM’s strategy, or that we don’t have time to pursue due to other constraints,” said Talamadupula. “Having the student teams work on them allows us to make meaningful progress without using internal resources. Also, we don’t want to get stuck in groupthink; we want fresh ideas infused.”