By Josh Baxt
Bodhisattwa (Bodhi) Prasad Majumder, a UC San Diego Department of Computer Science and Engineering (CSE) PhD student, has been honored with an Adobe Research Fellowship. The fellowships are only awarded to about ten people each year, and Majumder is the third CSE graduate student to receive one, following Ailie Fraser in 2017 and Zexiang Xu in 2019.
Majumder’s focus for the Adobe project is explainable interactive systems. Consider Netflix, which makes recommendations and explains them based on previous viewing habits. However, some of the suggestions don’t make a lot of sense, and there are no mechanisms to correct them. This is true for many of these systems.
“These user interfaces are now quite static and not always trustworthy,” said Majumder. “They can be quite frustrating. You may be dissatisfied with the results and want to speak with an actual human, if you have that option. But you should be able to critique them and say: ‘Hey, I don't like your explanation and this is my reason.’ The system would take that feedback and provide a new prediction.”
Whether it’s watching Netflix, querying a tax service or ordering pizza, imbuing these systems with a little common sense could make them more intuitive, provide better information for users and generally help them earn the public’s trust. In some respects, they are like human toddlers, who don’t really understand the world around them.
“There are certain things we take for granted,” said Majumder, who is part of CSE Professor Julian McAuley’s group. “Ice is cold or don’t leave the refrigerator open, just common sense. I want to give this type of knowledge to the system, so that when it provides an explanation, it should make sense – not just in a digital way but in a real world scenario.”
Beyond Consumer Systems
These kinds of systems are becoming ubiquitous – Alexa, Siri, online shopping, customer service – but Majumder envisions even more concrete applications in healthcare and education. A medical triage system, for example, could help patients describe their symptoms in more meaningful ways.
“This could be quite useful for geriatric patients,” said Majumder. “Sometimes they can have difficulty providing detailed information about pain. The system could help elicit more precise descriptions to help clinicians personalize care.”
In education, explainable interactive systems could help make universities more user-friendly. Sometimes students are overloaded with information about their class choices, and they have trouble deciding which ways to go. A personal teaching assistant could help each student sort through the information and make the best decisions.
The Adobe Fellowship awards him $10,000, a one-year Creative Cloud subscription and the opportunity to intern at Adobe. This is the latest honor for Majumder, who received a Qualcomm Innovation Fellowship, along with fellow PhD student Shuyang Li, in 2020.