Research

We develop a computational model that explains how people make causal judgments in physical scenes by mentally simulating counterfactual outcomes and comparing those to what actually happened.

We aim to develop a systematic framework for robots to build models of the world and to use these to make effective and safe choices of actions to take in complex scenarios.

AIRvatar is a system that telemetrically collects and analyzes fine-grained data on users’ virtual identities and the process used to create them.

Alloy is a language for describing structures and a tool for exploring them. It has been used in a wide range of applications from finding holes in security mechanisms to designing telephone switching networks. Hundreds of projects have used Alloy for design analysis, for verification, for simulation, and as a backend for many other kinds of analysis and synthesis tools, and Alloy is currently being taught in courses worldwide.

Robot scheduling via learning from demonstration for tasks in medicine and more

In order to be able to design synthetic organs that function autonomously, we will need to engineer artificial tissue homeostasis. To control the size of these artificial tissues, two major mechanisms will have to be engineered.

Automatic speech recognition (ASR) has been a grand challenge machine learning problem for decades. Our ongoing research in this area examines the use of deep learning models for distant and noisy recording conditions, multilingual, and low-resource scenarios.

Adding domain knowledge to word embeddings.

We study the fundamentals of Bayesian optimization and develop efficient Bayesian optimization methods for global optimization of expensive black-box functions originated from a range of different applications.

We develop image analysis algorithms for whole-heart cardiac MRI in patients with severe congenital heart disease.

We aim to study the impact of computer-supported roleplaying in changing social perspectives of digital media users. Such media could take the form of videogames, VR systems, training software, and other types of interactive narrative technology.

We develop machine learning algorithms to automatically quantify the severity of pulmonary edema from chest x-rays.

Integrating multi-modal clinical data and using recurrent and convolution neural networks to predict when patients will need important interventions.

We develop novel algorithms for neuroimage segmentation, registration, and joint modeling of brain structure with clinical and genetic data.

Our focus is on improving health-related outcomes by creating machine learning models that improve human risks.

Our goal is to develop new applications and algorithms that leverage the skills of distributed crowdworkers, notably for image and video processing applications.

Mixed-methods qualitative (interviews and coding) and computational (AI) approach to understanding relationships between social identities, cultural values, and virtual identity technologies (e.g., online profiles and avatars).

Our team uses accelerometers and machine learning to help detect vocal disorders. We capture data about the motions of patient's vocal folds to determine if their vocal behavior is normal or abnormal.

The robot garden provides an aesthetically pleasing educational platform that can visualize computer science concepts and encourage young students to pursue programming and robotics.

We use computational models of biological pathways in disease to estimate the range of possible responses to various hypothetical drugs, given patient variability and limited pathway-relevant data.

EQ-Radio can infer a person’s emotions using wireless signals. It transmits an RF signal and analyzes its reflections off a person’s body to recognize his emotional state (happy, sad, etc.).

Existing methods for cloning and recombination of DNA enable construction of arbitrary sequences. However, the sequential nature of these techniques makes them time-consuming and expensive. Furthermore, while the transformation of an existing plasmid into a host strain can be reliable when a selection marker is used, there are many current limitations: the number of different plasmids that can be co-transformed is limited by the choice of markers and compatible origins of replication; plasmids are less stable than chromosomal DNA and are difficult to maintain indefinitely without mutation; and cistronic interactions cannot be designed since each new nucleotide sequence added is on an unconnected DNA molecule. To overcome these limitations, we are designing reconfigurable chromosomes consisting of both fixed and variable regions. While the fixed region is carefully optimized and tuned ahead of time, the variable region can be modified in the field, at the point-of-use, leading to rapid and on-demand realization of novel biocircuits with many different phenotypes.

Our goal is to help medical researchers and clincians understand the growing repositories of waveform and signal data collected from critically ill patients.

From audio recordings of clinician-subject interactions, we determine the spoken language bio-markers that are associated with health outcomes, such as dementia and depression.