MIT-IBM Watson AI Lab Seed Demonstration: Maximizing the Impact of Early Career Skills | MIT News

The early years of faculty members’ careers are a formative and exciting time to establish a solid footing that helps determine the trajectory of researchers’ studies. This involves building a research team, which requires new ideas and guidance, creative collaborators, and reliable resources.

For MIT’s artificial intelligence division, early collaboration with the MIT-IBM Watson AI Lab on projects played a key role in helping to inspire ambitious lines of research and shaping multiple research teams.

Building momentum

“The MIT-IBM Watson AI Lab was very important to my success, especially when I was just starting out,” said Jacob Andreas – an associate professor in the Department of Electrical Engineering and Computer Science (EECS), a member of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), and an MIT-IBM Watson language researcher – who studies Lab AI. Shortly after joining MIT, Andreas started his first major project using the MIT-IBM Watson AI Lab, working on language representation and structured methods for adding data to less commonly used languages. “It was definitely something that made me launch my lab and start hiring students.”

Andreas notes that this happened at a “crucial time” when the field of NLP was evolving to understand language models – a task that required more computing, which was available through the MIT-IBM Watson AI Lab. “I feel like the kind of work we did under that [first] project, and working with all of our people on the IBM side, was very helpful in finding a way to navigate that transition.” In addition, the Andreas group was able to pursue multi-year projects in pre-training, reinforcement learning, and measurement of reliable responses, thanks to the computing resources and technology within the MIT-IBM community.

For several other faculty members, timely participation with the MIT-IBM Watson AI Lab proved to be very beneficial as well. “Having the intellectual support and being able to use some of the resources within MIT-IBM, that has been completely transformative and very important to my research program,” said Yoon Kim – associate professor in EECS, CSAIL, and MIT-IBM Watson AI Lab researcher – who also saw the field of his research. Before joining MIT, Kim met his future collaborators during an MIT-IBM postdoctoral fellowship, where he pursued the development of a neuro-symbolic model; Now, Kim’s team is developing methods to improve the master language model (LLM) skills and efficiency.

Another factor he points to that has led to his team’s success is a seamless research program with smart colleagues. This has allowed his MIT-IBM team to apply the project, test it at scale, identify issues, validate strategies, and adapt requirements to develop shortcuts for possible implementation in real-world applications. “This is the impetus for new ideas, and I think that’s what’s unique about this relationship,” Kim said.

Integrating technology

The nature of the MIT-IBM Watson AI Lab is that it not only brings together researchers in the area of ​​AI to accelerate research, but also integrates work across disciplines. Lab researcher and MIT associate professor in EECS and CSAIL Justin Solomon describes his research team as having grown through the lab, and the collaboration as “important … from the beginning to now.” Solomon’s research team focuses on theoretically oriented, geometric problems as they relate to computer graphics, vision, and machine learning.

Solomon credits the MIT-IBM collaboration with expanding his skill set and the application of his team’s work — sentiments echoed by lab researchers Chuchu Fan, associate professor of aerospace engineering and member of the Information Systems and Decisions Laboratory, and Faez Ahmed, professor of mechanical engineering. “See [IBM] they are able to translate some of these complex problems from engineering to the kind of mathematical tools that our team can work on, and close the loop.” This, for Solomon, involves combining different AI models trained on different datasets to perform different tasks. “I think these are all really exciting places,” he said.

“I think these are early applications [with the MIT-IBM Watson AI Lab] It’s very much shaping my research agenda,” said Fan, whose research intersects with robotics, control theory, and security critical systems. Like Kim, Solomon, and Andreas, Fan and Ahmed started projects by working together the first year they were able to do it at MIT. Constraints and efficiency dominate the problems facing Fan and Ahmed, so they require deep domain knowledge outside of AI.

Collaboration with the MIT-IBM Watson AI Lab has enabled the research team to combine formal methods with natural language processing, which, he says, has allowed the team to move from developing automated tasks and motion planning for robots to building LLM-based agents for travel planning, decision making, and validation. “That work was the first experiment of using LLM to translate any kind of free natural language into something that a robot can understand, that can use. That’s something I’m proud of, and very difficult at the time,” Fan said. In addition, through joint research, his team was able to develop an LLM concept – work that “would not have been possible without the support of IBM,” he said.

In the lab, Faez Ahmed’s collaboration has facilitated the development of machine learning methods to accelerate discovery and design within complex mechanical systems. Their Communications function, for example, uses “generative optimization” to solve engineering problems in a data-driven and precise manner; more recently, they are using multidimensional data and LLMs in computer aided design. Ahmed says AI is often used for problems that have already been solved, but could benefit from increased speed or efficiency; however, challenges – such as mechanical communication that were considered “almost unsolvable” – are now within reach. “I think that’s a sign [of our MIT-IBM team],” Ahmed said, praising the success of his MIT-IBM team, co-led by IBM’s Akash Srivastava and Dan Gutfreund.

What began as an initial interaction for each MIT faculty member has evolved into a lasting faculty relationship, where both parties are “excited about science,” and “student-driven,” adds Ahmed. Taken together, the experiences of Jacob Andreas, Yoon Kim, Justin Solomon, Chuchu Fan, and Faez Ahmed speak to the impact that strong, active, academic-industry relationships can have in establishing research teams and engaging scientific exploration.