The Computing Research Association (CRA) is a coalition of more than 200 organizations with the mission of enhancing innovation by joining with industry, government, and academia to strengthen research and advance education in computing. Every year, they recognize North American students who show phenomenal research potential with their Outstanding Undergraduate Researcher Award, and in 2021, Brown CS made its strongest showing yet. Nishanth Kumar, a Computer Engineering concentrator and Brown CS researcher, is one of 23 Finalists, and out of 73 students who received Honorable Mentions, three of them are Brown CS students: Sarah Bawabe, Dylan Sam, and Homer Walke. Last year, Brown CS students received four Honorable Mentions.
Asked to situate his work, Nishanth explains that overall, it sits at the intersection of AI and robotics and seeks to enable robots to solve useful tasks for humans in the real world: "To illustrate a recent project, let's try a quick and simple thought experiment. Suppose you find yourself in a kitchen you've never been in before and would like to brew yourself some coffee. You might start looking for a few relevant things, like a cup, a coffee maker, some milk, etc. But there's a litany of irrelevant things – such as whether or not it is raining outside or how many empty soda bottles are in the kitchen fridge – you don't even bother to consider. More generally, we humans possess large amounts of knowledge about many different things in the world, yet when confronted with a specific task, we're able to immediately direct our focus to only those few things that are relevant to achieving our goals. My recent work under Professors Stefanie Tellex, George Konidaris, and Michael Littman has attempted to imbue AI agents with similar capabilities."
"One of the biggest questions," says Sarah Bawabe, "that we have asked in Professor Jeff Huang’s User Interfaces and User Experience course is: 'How can we make an inherently subjective grading process more objective?' Listening to students' complaints of potentially subjective grading methods in the past, I chose to help Jeff tackle this issue by creating a new grading platform built upon peer-to-peer assessments. This web application, titled The UX Factor, allows students the ability to look at their classmates’ work anonymously and provide guided feedback. Students would be shown two of their peers’ submissions and answer comparative questions chosen by instructors. Through building this application, we have been able to pioneer a new approach both to grading and to teaching and promoting the process of critique. Although this application was built specifically with the User Interfaces and User Experience course in mind, it has already been utilized to help our TA staff give feedback to each others’ edited course materials, and it is looking to be extended to other courses to help them reduce their TA grading workload. We are hoping in the future to continue to see new applications for this platform both in other classrooms and in other fields where critiquing and feedback are critical to an efficient workflow."
Dylan explains that his research studies weakly supervised learning from a statistical and theoretical perspective. "Weakly supervised learning," he says, "looks to reduce the necessity of large amounts of labeled data through the usage of weak labelers, which are either hand-engineered heuristics or noisy classifiers trained on related data. Most recent advances in weakly supervised learning assume that the predictions of these weak labelers are independent. However, this assumption is not realistic and frequently violated, and attempts to learn without this assumption are purely empirical and do not provide mathematical guarantees. Over the past years, I have been incredibly fortunate to work with Alessio Mazzetto, Cyrus Cousins, and Professors Stephen Bach and Eli Upfal to develop the first theoretically justified algorithm for learning from weak labelers, without any assumptions on their distributions."
"My work with Professor Littman," says Homer, "has focused on producing linear temporal logic (LTL) formulas from demonstrations. In robotics, LTL is used as a formal specification of tasks. Given human demonstrations of a behavior (for example, navigating a room or stacking a series of blocks), our method produces an LTL formula that summarizes the behavior. A robot could then use the formula as a guide to perform the task. My work with Professor Ritchie has addressed unsupervised visual program induction. In this problem, the goal is to infer programs that, when executed, reconstruct a set of target shapes. We designed a method that learns to infer visual programs for shapes, even without access to labeled training data. These visual programs enable easy editing of the shape and are useful in multiple domains that require 2D or 3D graphics, from video games to visual effects for movies."
The full list of Outstanding Undergraduate Researcher Award recipients is available here.
For more information, click the link that follows to contact Brown CS Communication Outreach Specialist Jesse C. Polhemus.