Artificial intelligence research is constantly developing fresh hypotheses that have the potential to benefit society and industry; however, sometimes these benefits are not fully realized due to lack of engineering tools. To aid bridge this gap, MIT’s 6-A Master of Engineering (MEng) students work with some of the world’s most novel companies and collaborate on cutting-edge projects while contributing to and completing their MEng thesis.
For part of last year, four 6-A MEng students partnered and interned with IBM Research’s advanced prototyping team through the MIT-IBM Watson AI Lab for AI projects, often developing web applications to solve real-world problems or business apply cases. Here, students worked with artificial intelligence engineers, user experience engineers, end-to-end researchers, and generalists to address design requests and receive thesis advice, says Lee Martie, an IBM research associate and 6-A manager. Student projects ranged from generating synthetic data to enable analysis of privacy-sensitive data to using computer vision to identify activities in video to monitor human safety and track construction progress on a construction site.
“I appreciate all the team’s knowledge and feedback,” says 6-A alumna Violetta Jusiega ’21, who participated in the program. “I think working in industry gives you the confidence that the needs of the project will be met and [provides the opportunity] solidify the research and make sure it will be helpful in some future apply cases.
Jusiegi’s research spanned the areas of computer vision and design, focusing on data visualization and medical user interfaces. Working with IBM, it built an application programming interface (API) that allows clinicians to interact with an AI model of treatment strategies that was deployed in the cloud. Its interface provided a medical decision tree as well as some recommended treatment plans. After receiving feedback on her design from physicians at her local hospital, Jusiega iterated on the API and a way to visually display results to make it user-friendly and understandable to clinicians who typically don’t code. He says that “these tools are often not used in the field because they lack some of the API principles that become more important in an industry where everything happens very quickly anyway, so there is little time to implement new technology.” However, this project may eventually allow for industrial implementation. “I think this application has great potential, whether it is used by clinicians or simply used in research. “It’s very promising and very exciting to see how technology can help us modify or improve the field of healthcare to be even more tailored to the needs of patients and provide them with the best possible care,” he says.
Another 6-A graduate, Spencer Compton, also considered helping professionals make more informed decisions for apply in settings including health care, but approached it from a causal perspective. Given a set of related variables, Compton investigated whether there was a way to determine not only the correlation, but also the cause and effect relationship between them (the direction of interaction) from the data itself. To do this, he and his colleagues at IBM Research and Purdue University turned to a field of mathematics called information theory. Aiming to design an algorithm to learn convoluted networks of causal relationships, Compton used concepts related to entropy, or randomness, in a system to aid determine whether a causal relationship existed and how variables might interact. “When evaluating explanations, people often resort to Occam’s razor,” Compton says. “We are more likely to believe in a simpler explanation than in a more complex one.” In many cases, he says, it seemed to work well. For example, they could have taken into account variables such as lung cancer, pollution and X-ray results. He was pleased that his research had allowed him to create a framework of “entropic causal inference” that could satisfactorily aid make sheltered and wise decisions in the future. “The math is really surprisingly deep and interesting and complex,” Compton says. “We’re basically asking, ‘When is the simplest explanation correct?’ but as a math question.
Determining relationships in data can sometimes require gigantic volumes of data to detect patterns, but for data that may contain sensitive information, this may not be available. For her master’s thesis, Ivy Huang worked with IBM Research to generate synthetic tabular data using a natural language processing tool called a transformer model, which can learn and predict future values from past values. Trained on real data, the model can generate fresh data with similar patterns, properties, and relationships without the limitations such as privacy, availability, and access that may be associated with real data in financial transactions and electronic health records. Additionally, it created an API and deployed the model to an IBM cluster, which allowed users to have greater access to and query the model without disturbing the original data.
Working with the Advanced Prototyping team, engineering candidate Brandon Perez also wondered how to collect and examine data with constraints, but in his case it was the apply of computer vision frameworks, focused on an action recognition model, to identify events in the field construction. The team based their work on the Moments in Time dataset, which contains over a million three-second video clips with about 300 classification labels attached, and performed well in AI training. However, the group needed more construct-based video data. For this purpose, they used YouTube-8M. Perez has created a platform for testing and tuning existing object detection and action recognition models that can be plugged into a tool for automatic spatial and temporal localization – how they identify and tag specific actions on a video timeline. “I was satisfied that I could explore what interested me, and I was grateful for the autonomy I was given in this project,” says Perez. “I felt like I was always supported and my mentor was a huge supporter of the project.”
“The type of collaboration we saw between our MEng students and IBM researchers is exactly what the 6-A MEng Thesis program at MIT is all about,” says Tomas Palacios, professor of electrical engineering and department director of the MIT 6-A MEng Thesis Program . “For more than 100 years, 6-A has connected MIT students with industry to solve some of the world’s most important problems.”